10 Krey Boulevard Rensselaer, NY 12144

September 21, 2012

By Electronic Delivery

Kimberly D. Bose Secretary

Federal Energy Regulatory Commission 888 First Street, N.E.

Washington, D.C. 20426

Re: Joint Study Commissioned by ISO RTO Council and Submitted in

Compliance with Federal Energy Regulatory Commission’s Order No. 745
Requiring the RTOs/ISOs to Examine the Requirements for and Impacts of
Implementing a Dynamic Net Benefits Approach to the Dispatch of
Demand Resources in the Wholesale Energy Markets; Docket RM10-17-
000

Dear Secretary Bose:

The ISO RTO Council (“IRC”) respectfully submits the attached report entitled Options
for Implementing a Dynamic Net Benefits Test Based on the Billing Unit Effect (“IRC
Report”) in compliance with the Federal Energy Regulatory Commission’s Order No.
745.1 The IRC is an industry organization that originally formed in the mid-1990s to
support the introduction of competition and open access transmission service in wholesale
power markets. It is now comprised of the 10 current North American ISOs and RTOs
including Alberta Electric System Operator (“AESO”), California Independent System
Operator (“CAISO”), Electric Reliability Council of Texas (“ERCOT”), Ontario’s
Independent Electricity System Operator (“IESO”), ISO New England (“ISO-NE”),
Midwest Independent Transmission System Operator (“MISO”), New York Independent
System Operator (“NYISO”), New Brunswick System Operator (“NBSO”), PJM
Interconnection (“PJM”) and Southwest Power Pool (“SPP”). The IRC works in a
collaborative fashion to develop effective tools standards, protocols and procedures to
improve competitive energy markets across North America.

1 Demand Response Compensation in Organized Wholesale Energy Markets, Order No. 745, FERC Stats. & Regs. 31, 322 (2011) (hereafter, “Order No. 745”). Paragraph 84 of Order No. 745

The IRC members2 appreciate the opportunity the Commission has provided to present a collective examination of the requirements, and impacts of, implementing a dynamic net benefits test in the dispatch of demand resources in day-ahead and real-time energy
markets. The IRC Report, in conjunction with individual compliance filings, is intended to meet the compliance filings obligations for the CAISO, ISO-NE, MISO, NYISO, PJM and SPP. The individual filings will incorporate by reference this IRC Report and will
highlight and expand upon the pertinent analysis and discussion presented in the IRC
Report as applicable to the commitment and dispatch software and algorithms that are
unique to each individual ISO’s or RTO’s markets.3

The IRC Report, which was prepared by Dr. Scott Harvey in collaboration with the IRC
Members, examines the range of implementation choices, issues, and potential costs
associated with the implementation of a dynamic net benefits calculation based on the
billing unit effect as a trigger for the activation of economic demand response. A high level
description of the operation of ISO/RTOs’ unit commitment and dispatch software and
multi-settlement systems is the starting point for this examination. Implementation issues
associated with application of a dynamic net benefits calculation based on the billing unit
effect in ISO and RTO unit commitment and dispatch software are then discussed in
conjunction with an evaluation of four alternative approaches to implementing such a
dynamic net benefits test based on the billing unit effect in ISO and RTO day-ahead and
real-time energy markets. The advantages and disadvantages in implementing of each
alternative are presented. The four alternative approaches examined in the paper are:

• Attempt to develop a solution to the unit commitment and dispatch problem that
applies a net benefits test based on the billing unit effect, utilizing known
mathematical dual optimization techniques and equilibrium constraints;

• Attempt to develop new solution concepts that might permit a faster and better unit
commitment and dispatch solutions to applying a net benefits test based on the
billing unit effect;

• Apply an ad hoc approach to apply a net benefits test based on the billing unit

effect utilizing existing software solution methods that would allow an evaluation of the billing unit effect based on making all demand response bids available for dispatch versus no demand response bids available for dispatch; and

• Apply an ad hoc approach to apply a net benefits test based on the billing unit

effect utilizing existing software solution methods that would allow an evaluation of the billing unit effect from making groupings of demand response bids available for dispatch.

2 The IESO, AESO, and NBSO are not subject to the Commission’s jurisdiction and did not participate

directly in the development of the IRC Report . In addition, ERCOT, which is not subject to the compliance obligations the Commission set forth in Order No. 745, did work collaboratively with the remaining members to support the development of the IRC Report.

3 Each ISO and RTO relies on somewhat different software and processes to schedule and dispatch resources in their day-ahead and real-time energy markets, but all ISOs and RTOs coordinating such markets use
software engines having four distinct components: 1) a unit commitment process; 2) a dispatch engine; 3) a powerflow calculation; and 4) a price calculation step.

The basic conclusion of the IRC Report is that the implementation of a more dynamic net
benefit test in the dispatch would likely have adverse impacts on the solution time of ISO
and RTO real-time dispatch software and day-ahead market software. To address these
impacts might likely require simplifications in other elements of the economic
commitment and dispatch software platforms that would result in less efficient solutions
and higher prices for power consumers. Further, the known software formulations utilized
by, and available to, ISOs and RTOs today would necessarily require such a net benefit test
to be implemented in a very restricted manner that would have the potential to routinely
produce anomalous market outcomes and still may not adequately achieve the objectives of
the dynamic net benefits test. The four alternative approaches examined would require, at
a minimum, substantial changes to the existing ISO/RTO software and in some cases may
require entirely new commitment and dispatch algorithms and software.

The IRC respectfully requests that the Commission accept this joint submittal, in

conjunction with the individually filed compliance filings of CAISO, ISO-NE, MISO,

NYISO, PJM and SPP, to meet the compliance obligations set forth by the Commission in paragraph 84 of Order No. 745.

Respectfully Submitted,

Nancy SaracinoStephen G. Kozey*

General CounselVice President, General Counsel, and

Sidney Davies Assistant General CounselSecretary

John Anders*Midwest Independent Transmission

Senior CounselSystem Operator, Inc.

California Independent System OperatorP.O. Box 4202

CorporationCarmel, Indiana 46082-4202

250 Outcropping Wayskozey@midwestiso.org

Folsom, California 95630

janders@caiso.com

Matthew Morais*Carl F. Patka*

Assistant General CounselAssistant General Counsel

Electric Reliability Council of Texas, Inc.Raymond A. Stalter

2705 West Lake DriveDirector of Regulatory Affairs

Taylor, Texas 76574New York Independent System Operator,

mmorais@ercot.comInc.

10 Krey Blvd

Rensselaer, New York 12144

cpatka@nyiso.com

Raymond W. HepperCraig Glazer*

Vice President, General Counsel, andVice President-Federal Government Policy

SecretaryPJM Interconnection, L.L.C.

Theodore J. Paradise*Suite 600

Assistant General Counsel, Operations and1200 G Street, N.W.

PlanningWashington, D.C. 20005

ISO New England Inc.202-423-4743

One Sullivan Roadglazec@pjm.com

Holyoke, Massachusetts 01040

tparadise@iso-ne.com

Paul Suskie*

Senior Vice President, Regulatory Policy and General Counsel

Southwest Power Pool, Inc.

415 North McKinley, Suite 140
Little Rock, Arkansas 72205

psuskie@spp.org

* = persons designated to receive service

CERTIFICATE OF SERVICE

I hereby certify that I have this day served the foregoing document upon each person

designated on the official service list compiled by the Secretary in this proceeding in accordance
with the requirements of Rule 2010 of the Rules of Practice and Procedure, 18 C.F.R. §
385.2010.

Dated at Rensselaer, NY this 21st day of September, 2012.

By:/s/ John C. Cutting

John C. Cutting

Senior Regulatory Affairs Specialist

New York Independent System Operator, Inc.

10 Krey Blvd.

Rensselaer, NY 12144 (518) 356-7521

Options for Implementing a Dynamic Net Benefits Test
Based on the Billing Unit Effect

Prepared by Scott Harvey
September 5, 2012

Prepared for the ISO RTO Council

Executive Summary

FERC’s Order 745 instructed the ISOs and RTOs to examine “the requirements for, costs
of, and impacts of implementing a dynamic net benefits approach to the dispatch of
demand resources that takes into account the billing unit effect in the economic dispatch
in both the day-ahead and real-time energy markets.”1 This paper is one part of the ISOs
and RTOs response to the Commission’s inquiry. The basic finding of this paper is that
the only method of implementing a net benefit test based on the billing unit effect using
existing, known software formulations would implement the test in a very restricted
manner that would have the potential to routinely produce anomalous outcomes, and even
this restricted implementation would likely have adverse impacts on the solution time of
ISO and RTO real-time dispatch software and day-ahead market software. These
solution time impacts might in turn require simplifications in other elements of the
software that would result in less efficient solutions, higher cost of meeting load, and
higher prices for power consumers.

The discussion of issues associated with implementation of a dynamic net benefits test based on the billing unit effect is structured around an evaluation of four general
approaches that could be taken to implementing such a dynamic test in either the dayahead or real-time energy markets. The four approaches are:

• Attempt to develop new solution concepts that might permit a faster and better
unit commitment and dispatch solutions to applying a net benefits test based on
the billing unit effect;

• Apply an ad hoc approach to apply a net benefits test based on the billing unit

effect utilizing existing software solution methods that would allow an evaluation of the billing unit effect from making groupings of demand response bids
available for dispatch.2

The first two approaches entail developing rigorous mathematical solutions to implement
a net benefits test based on the billing unit effect. Either or both approaches might
ultimately yield high quality solution concepts that could be implemented in commercial
programs that would apply a net benefits test based on the billing unit effect in on line
operation. However, both of these approaches require significant focused research efforts
whose outcome and timeline is uncertain. At the end of two or three years they might
produce the requirements to allow the development of programs that would yield high
quality solutions to the net benefit test based on the billing unit effect within solution
time frames that would be acceptable to ISOs and RTOs from an operating perspective,
or they might not.

Implementation of a net benefit test based on the billing unit effect utilizing the third
approach, on the other hand, would not require the development of any new solution
concepts or methods, so could provide the basis for ISOs and RTOs to move forward in
the near-term with the development of software to implement a dynamic net benefits test
based on the billing units effect. This approach has two limitations, however, that the
ISO’s and RTO’s should be aware of. First, the essence of this approach is that it only
compares the billing unit effect benefits of making all demand response bids available for
dispatch or none. This has the consequence that some demand response bids that would
pass the billing unit effect based net benefits test on their own, could fail to pass the test
in combination with all other demand response bids, or conversely that bids that would
not pass on their own, might pass in combination with other more cost effective demand
response bids. This limitation of the third approach could have undesirable consequences
when the billing unit effect based net benefits test is applied in the real-time dispatch and
even more undesirable consequences when applied to the day-ahead market’s
optimization over the 24 hours of the operating day. Second, the increase in computation
time required to implement this approach would require compensating tradeoffs in terms
of extending software solution time or reducing other software functionality that would
tend to raise the cost of meeting load and/or adversely impact reliability or incurring the
costs required to solve two cases in parallel.3

The fourth approach would, similar to the third approach, attempt to utilize only current
software algorithms allowing a dynamic billing unit effect based net benefit test to be
implemented within a defined time frame. It would attempt to avoid the undesirable
consequences of the all or nothing test applied by the third approach, and apply a net
benefits test based on the billing unit effect to groups of demand response bids.
However, while there are ad hoc methods that could perhaps be used to apply this fourth
approach to the day-ahead or real-time dispatch on an uncongested system, these ad hoc
methods would not be workable when applied to a congested transmission system or to
the day-ahead unit commitment, unless combined either with fundamental changes in the
structure of the dispatch software or a research effort to attempt to develop algorithms or
solution methods that would provide an acceptable solution within an acceptable time

frame from an ISO or RTO operating perspective as under the first and second approaches.

Beyond the implementation challenges associated with each of these approaches,

implementation of a solution based on any of these approaches would require making a
variety of compromises in how a net benefits test based on the billing unit effect would
be applied:

• In real-time dispatch software that includes either full inter-temporal
optimization or look-ahead ramp management functionality;

• In real-time pricing systems that include special pricing rules such as ex

post pricing, ELMP pricing, fixed block pricing or separate scheduling and pricing passes;

• In demand response evaluations in market power mitigation passes or
intra-day look-ahead scheduling and unit commitment evaluations ;

• To evaluate the billing unit effect on the day-ahead market of real-time
price impacts of demand response activation;

These compromises are discussed in detail in Section IV. The critical problem in using
existing software programs, algorithms and solution methods to apply a net benefits test
based on the billing unit effect in a dynamic application is that the dispatch of the demand response resource would be contingent on the outcome of the net benefits test based on
the billing unit effect.

Several recent FERC orders suggest that the Commission does not intend that the

dispatch of demand response resources depend on the outcome of the net benefits test
based on the billing unit effect, rather, it appears that the Commission intends that
demand response resources be dispatched based on their bids, i.e. based on a
conventional production cost minimizing benefit test, with only the nature and amount of
their compensation potentially depending on the outcome of the net benefits test based on
the billing unit effect. If this understanding is correct, the dispatch of demand response
resources based on their bid could be implemented using existing software tools, and the
application of the net benefit test based on the billing unit effect could be carried out after
the fact in the settlements process, i.e. there would be no need for a “dynamic” net benefit
test based on the billing unit effect in the sense of a test carried out as part of the real-
time economic dispatch or in the process of clearing the day-ahead market. While the
application of the net benefit test based on the billing unit effect would be complex to
carry out even within an after the fact settlement process and would require some
simplifications, appropriate simplifications would permit it to be applied without the
development of new market software algorithms or solution concepts.

iii

Table of Contents

I.Overview..........................................................1

II. Design and Structure of Industry Dispatch and Unit Commitment Software..2

III. Net Benefits Test Implementation Choices and Issues...................4

A. Introduction.................................................4

B.Develop New Software Applying Known Methods......................9

C.Develop New Solution Methods; Develop New Software Applying These

Methods......................................................10

D. Modify Existing Software to Apply a Net Benefits Test Based on the Billing

Unit Effect on an All or Nothing Basis...............................10

E.Modify Existing Software to Apply a Net Benefits Test Based on the Billing

Unit Effect to Groups of Demand Response Bids......................15

F.Conclusions....................................................19

IV. Other Design Choices and Market Impacts..........................20

A. Interaction with real-time dispatch software that includes inter-temporal

optimization...................................................21

B.Interaction with special RTO pricing rules...........................24

C.Accounting for the real-time price reductions on the day-ahead supply curve.. 27

D. Interaction with market power mitigation processes................29

E.Impact on in day look-ahead scheduling and unit commitment evaluations..34

F.Interaction with Ramp Capability Products...........................37

V. Alternative Approaches..........................................38

VI. Recommendations..............................................40

Appendix A...........................................................42

Options for Implementing a Dynamic Net Benefits Test Based on the Billing Unit Effect4
Prepared by Scott Harvey5

September 5, 2012

I.Overview

Section II of this paper provides a high level description of the operation of industry unit
commitment and dispatch software and multi-settlement systems to provide background
and context for the later discussion. Section III describes the implementation issues that
would be associated with application of a dynamic net benefits calculation based on the
billing unit effect in ISO and RTO unit commitment and dispatch software, then
describes and evaluates the four alternative approaches to implementing such a dynamic
net benefits test based on the billing unit effect in ISO and RTO day-ahead markets and
real-time dispatch summarized above. Section IV discusses a number of secondary

ERCOT, ISO New England, MISO, the New York ISO, PJM LLC, and the Southwest Power Pool. The
IRC is comprised of the Alberta Electric System Operator (“AESO”), the California Independent System
Operator (“CAISO”), Electric Reliability Council of Texas (“ERCOT”), the Independent Electricity System
Operator of Ontario, Inc., (“IESO”), ISO New England, Inc. (“ISONE”), Midwest Independent
Transmission System Operator, Inc., (“Midwest ISO”), New Brunswick System Operator (“NBSO”), New
York Independent System Operator, Inc. (“NYISO”), PJM Interconnection, L.L.C. (“PJM”), and Southwest
Power Pool, Inc. (“SPP”). The AESO, IESO, and NBSO are not subject to the Commission’s jurisdiction,
and these comments do not constitute agreement or acknowledgement that they can be subject to the
Commission’s jurisdiction. ERCOT is not subject to the Commission’s jurisdiction with respect to the
issues presented in this NOPR, but is joining in support of the IRC comments. The IRC’s mission is to
work collaboratively to develop effective processes, tools, and standard methods for improving the
competitive electricity markets across North America. In fulfilling this mission, it is the IRC’s goal to
provide a perspective that balances Reliability Standards with market practices so that each complements
the other, thereby resulting in efficient, robust markets that provide competitive and reliable service to
customers.

issues in implementing a dynamic net benefits test based on the billing unit effect that

would impose limitations on the design and/or unit commitment and dispatch results that the Commission should be aware of in evaluating alternative approaches to implementing such a dynamic test.

Section V discusses FERC comments and instructions in several Order 745 Compliance
and Rehearing Orders which suggest that a dynamic net benefits test based on the billing
unit effect of the kind that gives rise to the implementation problems discussed in
sections III and IV is not what FERC envisions for Order 745 compliance and hence is
not necessary to comply with the Commissions intent in Order 745. If this understanding
is correct, an alternative approach to implementing Order 745 that would apply a net
benefits test based on the billing unit effect in the settlement process would be feasible.

Finally, section VI briefly summarizes the alternatives and their advantages and

disadvantages in implementing the dynamic net benefits test based on the billing unit
effect.

II. Design and Structure of Industry Dispatch and Unit Commitment Software

Each ISO and RTO relies on somewhat different software and processes to schedule and
dispatch resources in their day-ahead and real-time energy markets, but all ISOs and
RTOs coordinating such markets use software engines having four distinct components:

1) a unit commitment process; 2) a dispatch engine; 3) a powerflow calculation; and 4) a price calculation step.

These components are distinct and in the case of programs developed by some vendors,
the program may cycle through these components more than once in the process of
reaching a solution. The unit commitment process determines which units are on-line
and available for dispatch in a particular period in the dispatch step. This is a
fundamentally different optimization problem than the economic dispatch because unit
commitment has binary, i.e. 0, or 1 choice variables, i.e. the unit is either off line or on
line and significant costs are typically incurred to start an off-line unit and keep it on line
at minimum load. Moreover, the unit commitment decision has irreversible impacts
because of unit’s minimum up times and down times. More complex formulations of the
unit commitment problem, like the one used by California ISO, include combined cycles
or multi-stage generation units with transition time, transition path, and transition cost in
the unit commitment formulation adding more complexity to the solution algorithm
when such a unit is (or could be) in transition from one configuration to another.

The unit commitment is therefore determined based on individual resource’s start-up

costs, minimum load or no-load costs, transition costs and times, incremental energy

costs, minimum run time, minimum down time, and other resource characteristics such as
their cost of providing various types of ancillary services. Because unit commitment is
an integer variable problem, not a conventional linear programming problem, distinct
solution methods, which vary from vendor to vendor, are used to determine the unit
commitment.

The dispatch step determines the amount of energy that each on-line unit is to produce. In
this step, the dispatch engine dispatches the available generation to meet load at least cost
on a production cost basis, given the unit commitment, while taking account of resources’
inter-temporal constraints, transmission constraints, both pre and post-contingency, and
other requirements, such as ancillary service requirements.7 These dispatch engines may
be solved by traditional linear programming methods or in some cases using mixed

integer programming. In all of these programs the re-dispatch to solve transmission constraints is based on linear shift factors representing the sensitivity of the flow on a transmission branch or flowgate to power injections at a particular bus/node.

The third component is a powerflow step which calculates the line flows associated with
the dispatch solution to determine whether there are overloads of any monitored element
in any base case or contingency case. The various vendors use different methods to
calculate these flows, some methods entailing an AC powerflow, quasi AC powerflow
solution and others use DC powerflow solutions. Some vendors in some programs solve
a powerflow repeatedly as the program iterates to a solution and others do not. This
powerflow step calculates the transmission overloads, if any, to be solved in additional
unit commitment and/or dispatch steps, and includes criteria for determining when the
line flows are within an acceptable range so that no further iteration is required.

The fourth component is a price calculation step, i.e. the step in which LMP prices are
calculated. The price calculation step is carried out after the final unit commitment has
been determined, and after the dispatch step has resolved any transmission constraint
violations. In existing software designs, these prices are not used or even known in the
process of committing generation or dispatching generation to meet load at least cost, but
are determined at the very end of the program cycle. As will be discussed further in
Section IV, special rules that vary from RTO to RTO sometimes apply in this price
calculation step.

Another common characteristic of most US ISOs and RTOs is that they coordinate both
day-ahead and real-time markets.8 These markets are operated as multi-settlement
systems in which day-ahead market schedules are settled at day-ahead market prices and
deviations between real-time generation or consumption and day-ahead schedules are
settled at real-time prices. These day-ahead and real-time markets are linked in ISO and
RTO systems by the use of common transmission system models and constraints, 9 and at
the market participant level by market participants reflecting their expectations regarding
real-time conditions in their day-ahead market bids, offers and schedules.

III.Net Benefits Test Implementation Choices and Issues

A.Introduction

This section discusses the core implementation issues associated with the use of a

dynamic net benefits test based on the billing unit effect as prescribed by the Commission
in Order 745 to trigger dispatch of demand response resources in real-time operation or to
schedule them in a day-ahead market. As explained by the Commission in its order, the
meaning of a “dynamic” test in this context is a test that is dynamic in the sense that it is
implemented within the real-time dispatch or within the day-ahead market, because the
outcome of the test determines whether demand response would be activated and hence
what level of demand must be met by the real-time dispatch or the day-ahead market

schedules.10

These core implementation issues are discussed below in the context of four general

approaches that could be used to implement such a dynamic net benefits test based on the billing unit effect. The issues discussed in this section are viewed as “core” issues
because they concern the ability of the ISOs and RTOs to acquire commercial software capable of implementing such a net benefits test within the ISOs and RTOs’ dispatch and unit commitment processes. A number of other less central issues which concern various limitations on the accuracy of such a dynamic net benefits test based on the billing unit effect implemented using these approaches are discussed in section IV.

The fundamental consideration that drives the need for careful consideration of the

method to be used to implement a dynamic net benefits test based on the billing unit

effect in ISO and RTO unit commitment and dispatch software is that the net benefits

calculation based on the billing unit effect is not a standard algorithm such as those

currently used in electric industry dispatch software which maximize net benefits through
the application of algorithms based on production cost minimization. As described in
section II, existing dispatch software does not even calculate clearing prices until the
dispatch problem has been solved. Hence, existing software algorithms would need to be
modified to calculate clearing prices at earlier points in the process of solving the
dispatch or alternatively make use of the shadow prices generated in the solution process
to apply a net benefits test based on the billing unit effect. The real problem, however is
that the changes in software design required to implement a net benefits test based on the
billing unit effect are much more fundamental. A design, in which prices are used to
determine the dispatch, rather than the least cost dispatch determining shadow prices and
clearing prices, requires fundamental changes in the solution process, so existing
software designs and algorithms cannot be used to directly implement the net benefits
calculation.

A further consideration in implementing a dynamic net benefits test based on the billing
unit effect is that all U.S. ISOs and RTOs settle their spot energy markets based on
locational prices (either zonal or nodal). In these markets, power consumers pay, and

generators are paid, the price of power at their location which may differ substantially
between locations on the transmission system as a result of transmission congestion.
Hence, there is not a single supply curve for an ISO or RTO market, there is a supply
curve for incremental power at each location within the ISO or RTO, given the dispatch
of the entire ISO or RTO to meet load at all other locations. Moreover, in a market with
transmission congestion, dispatching demand response to reduce or minimize net
payments by power consumers or to reduce or minimize payments to generators is not
equivalent to dispatching demand response to reduce or minimize gross energy market
payments. This is because gross energy market payments by power consumers on a
congested transmission system equal payments to generators plus congestion rents, and
congestion rents flow directly (through the allocation of FTRs, CRRs, or ARRs) or
indirectly (through the allocation of FTR, TCC or CRR auction revenues) to power
consumers. Hence, applying a net benefits test based on the billing unit effect must
deviate even further from traditional production cost minimizing solution concepts and
algorithms on a congested transmission system.

The billing unit effect benefit to power consumers from the dispatch of demand response to depress spot energy prices would be measured in the presence of congestion by
calculating the reduction in the net energy market payments by remaining load,11 which could be compared to the payment to demand response providers to account for the
billing unit effect in applying the net benefits test to the dispatch of demand response when there is transmission congestion.

One possible outcome from Order 745 implementation is that most or all economic
demand response bids would be submitted in real-time and evaluated in the real-time
dispatch. If this turns out to be the case, it would not be necessary to apply the net
benefits test based on the billing unit effect in the day-ahead market and the impact of
real-time demand response on day-ahead market prices would be felt through the
submission of virtual supply bids and price capped physical load bids in the day-ahead
market. The implications of such an outcome are discussed at length in subsection 5 of
section IV below. Another possibility is that a material number of demand response bids
would be submitted in the day-ahead market and their activation would be need to be
determined by the application of a net benefits test based on the billing unit effect.

There are three features of the unit commitment in the day-ahead market that would make
the application of the net benefits test based on the billing unit effect a more intractable
problem in the day-ahead market than would be the case in the real-time dispatch. These
are 1) the fundamental properties of the unit commitment problem and the methods
currently used to solve it would further complicate development of new algorithms for
solving the unit commitment problem based on a net benefits test that accounts for the

billing unit effect;12 2) further complications introduced by the need to jointly solve the
day-ahead market over the 24 hour time horizon of the operating day; and 3) the potential
for the application of a net benefits test based on the billing unit effect within the unit
commitment process to depress clearing prices through uneconomic generation
commitment, rather than through the intended dispatch of demand response resources.

In Order 745 FERC directed the ISOs and RTOs to depart from the production cost

minimizing benefits test in the dispatch of demand response if the dispatch of the demand response sufficiently depresses energy market clearing prices,13 but has not directed that they depart from production cost minimization in making other unit commitment and
dispatch decisions that would also depress energy market clearing prices and hence
potentially give rise to a billing unit effect. Hence, there is a sense in which the FERC
order envisions two distinct net benefits tests being applied within the unit commitment process, one based on production cost minimization and one based on the billing unit
effect. These features are briefly discussed below to provide context for the discussion of alternative implementation approaches which follows.

The security constrained economic dispatch used by many ISOs and RTOs to meet load
in real-time and to determine day-ahead schedules given the unit commitment generally
has the properties of a conventional linear programming problem with shadow prices that
correspond to clearing prices.14 As noted in section II, however, day-ahead markets also
include a unit commitment step that determines which units can be dispatched in the
dispatch step. The unit commitment problem is a complex integer variable problem that
cannot be solved with conventional linear programming methods. The various U.S. ISOs
and RTOs and their software vendors use a variety of methodologies, including mixed
integer programming, to determine the unit commitment in their day-ahead market.

Not only are existing unit commitment algorithms based on production cost minimization
rather than payment minimization, the current solution algorithms for the unit
commitment step do not develop explicit shadow prices (as is the case in the linear
programming solution to the economic dispatch problem) that could be used as a starting
point for the application of payment minimization within the unit commitment step. This
lack of prices could be addressed by using prices calculated in each dispatch step to guide
the iteration in the unit commitment step and either formulating the objective function in
terms of prices or imposing a side constraint that accounts for price effects, but these
prices will not provide a direct link between changes in the unit commitment and changes
in prices and net payments by remaining load as is feasible with a production cost
minimization objective function.

The second complication in determining the unit commitment is that it must be solved

jointly over the hours of the day to achieve the overall least cost unit commitment,

because of start-up costs, transition costs, minimum run-times, minimum load costs, ramp rates, forbidden regions, and minimum down times. Hence, all US ISOs and RTOs
jointly optimize their day-ahead markets over at least a 24-hour time frame. This
optimization over time in day-ahead markets has the consequence that a change in load in one hour can impact the unit commitment in a way that changes clearing prices in other hours. For example, in analyzing potential approaches to complying with Order 745 the NYISO reran a day-ahead market case with additional price capped load bids and found that the demand reduction in one hour attributable to the activation of a price capped load bid in that hour caused the clearing price to rise in the following hour. The objective of
the optimization is to minimize the sum or the total cost across all hours rather than
minimizing the cost of meeting load in each individual hour.

Third, it needs to be kept in mind that clearing prices could also be artificially depressed
by committing excess generation in the unit commitment process and paying more than
the clearing price for the generation’s output (i.e. paying uplift in addition to the LMP
price to cover the start-up and minimum load costs of the excess generation). Order 745,
however, only orders above market payments and assignment of uplift costs to consumers
for the uneconomic dispatch of demand response resources that sufficiently depresses
clearing prices in ISO and RTO markets. Hence, any logic structure that is developed for
use in implementing a dynamic net benefits test based on the billing units effect within
the unit commitment step would need to be structured so that it would dispatch demand
response to depress clearing prices based on the billing unit effect net benefits criteria
(whether implemented solely through changes in the objective function or also through
incorporation of sides constraints), but not uneconomically commit generation to depress
clearing prices.

Maintaining such a distinction without introducing unintended consequences would be
one of the additional challenges in applying a net benefits test based on the billing unit
effect within ISO and RTO unit commitment processes. When the dispatch of demand
response depresses clearing prices in the day-ahead market in particular, its dispatch
would potentially also make the commitment of some generation uneconomic. The
normal operation of ISO and RTO unit commitment software would then be to decommit
the uneconomic generation that is no longer needed or economic to meet demand.
However, decomitting uneconomic generation would reduce the price impact of
dispatching the demand response and perhaps cause demand response activation to fail
the net benefits test based on the billing unit effect as the change in the unit commitment
would replace low cost generation with higher cost demand response with little impact on
clearing prices.

If, however, an attempt were made to modify the unit commitment process to not

decommit uneconomic generation when demand response was activated so that demand
response activation would be more likely to artificially depress clearing prices, there
would be a potential for the unit commitment to depress clearing prices by simply
committing uneconomic generation, which would have a billing unit effect similar to the

activation of demand response. One of the complications in avoiding such an outcome is
that the commitment of excess generation (i.e. more than would be economic based on a
production cost minimizing benefit test) that depresses clearing prices would always pass
a net benefits test based on the billing unit effect, which does not consider the uplift costs
that would be associated with an inefficient unit commitment but does account for the
uplift costs associated with above market payments to demand response. Hence, there
might be a potential for day-ahead market solutions in which no demand response is

activated because clearing prices are too low, generation profits are reduced because

clearing prices are artificially depressed, but consumer costs rise because too many units are on line and generators receive substantial above market uplift payments.15

With these considerations in mind, we describe four possible approaches to implementing
a dynamic net benefits calculation based on the billing unit effect, discuss the
implementation issues, general timeline and costs associated with each, and identify
known issues with the quality of the solutions each approach would be capable of
producing.

One possible approach would be to develop a solution to the unit commitment and

dispatch problem that applies a net benefits test based on the billing unit effect utilizing known dual optimization techniques;

A second possible approach would be to attempt to develop new solution concepts that might permit a faster and better unit commitment and dispatch solutions applying a net benefits test based on the billing unit effect;

A third possible approach would be an ad hoc methodology that would apply a net

benefits test based on the billing unit effect utilizing existing software solution methods
that would allow an evaluation of the billing unit effect based on making all demand
response bids available for dispatch versus no demand response bids available for
dispatch;

A fourth possible approach would be to develop an ad hoc methodology that would apply a net benefits test based on the billing unit effect utilizing existing software solution
methods that would allow an evaluation of the billing unit effect from making groups of demand response bids available for dispatch.

Each of these general approaches is discussed below, describing each potential approach
in more detail, discussing in general terms the likely time frame in which ISOs and RTOs
would have the ability to acquire commercial software that would implement a net
benefits test based on the billing unit effect in the ISO’s and RTO’s unit commitment and
dispatch software, and commenting on known properties of the solutions produced by
each approach.

B.Develop New Software Applying Known Methods

As outlined above, the first possible approach to carrying out a dynamic net benefits test
based on the billing unit effect would be to develop a solution to the unit commitment
and dispatch problem that applies a net benefits test based on the billing unit effect
utilizing known dual optimization techniques. For instance, the net benefits test based on
the billing unit effect could theoretically be incorporated into the ISO and RTO’s
optimization as a new non-linear constraint under the current production cost
minimization objective function, or replace the current production cost minimization
objective function with a load or billing unit effect cost minimization objective function.
Both of these mathematical formulations are theoretically possible. The issue with this
approach is that both formulations would have a net benefit test expression that has cross-
product term of the bid megawatt dispatch variables in the corresponding constraint or in
the objective function that may call for the need of non-linear optimization solution
algorithms. It should also be noted that in a net benefits test formulation, the locational
prices are no longer byproducts of the optimization problem as they are in the current
approach and calculated as post process after the optimization. The clearing prices would
now be part of the optimization formulation.

The resultant net benefits test formulation is described as a self-referential Mixed Integer
Non-Linear Program problem. This type of problem formulation is known in
mathematics to be extremely challenging to solve efficiently, let alone optimally, and
requires long solution times due to its non-convexity, non-linearity, discreteness, and also
due to the less-developed mathematical techniques required to handle such mathematical
programs with difficult equilibrium constraints. Both the California ISO and ISO New
England have previously provided FERC with detailed discussions of the complexity of
solving this class of problem. 16 Both ISO New England and the California ISO have
observed that there is no commercially available program capable of solving large scale
problems of this type, let alone solving this class of problem within the solution time
frame of ISO and RTO day-ahead markets or real-time dispatch.

A particularly important unknown is whether these suggested solution methods could be
used or similar concepts used to apply a net benefits test based on the billing unit effect to
the solution of the unit commitment problem, rather than simply to the economic
dispatch. Beyond the unknown workability of applying these known solution methods to
the unit commitment problem, a further deeper unknown in pursuing this approach is the
workability of these solution concepts in solving the unit commitment problem if a net
benefits test based on the billing unit effect is applied to some choice variables (the
amount of demand response activation) and a net benefits test based on production cost
minimization to other choice variations (generation commitment and dispatch) within a
single unit commitment problem.

The solution concepts needed to implement a dynamic net benefits test based on the
billing unit effect would need to be developed, interactions with other elements of the

dispatch software resolved, and execution time impacts addressed, before a dynamic net
benefits test based on the billing unit effect and utilizing these prospective methods and
algorithms could be implemented. There cannot be a fixed timeline for implementation
of a dynamic net benefits test based the billing unit effect utilizing solution concepts and
algorithms that have yet to be developed. Moreover, until the algorithms that would be
used to implement the approach have been developed, evaluated, and tested, there cannot
be any assessment of any potential undesirable features that might be required to

implement such an approach.

C.Develop New Solution Methods; Develop New Software Applying These

New Methods.

A second approach that ISOs and RTOs could take to implementing a dynamic net

benefits test based on the billing unit effect would be to fund academic and/or industry
research that would attempt to develop new solution concepts that might permit a faster
and better application of a net benefits test based on the billing unit effect to both the unit
commitment and dispatch problems. This approach could focus additional research
specifically on the unit commitment aspects of the problem and explore the consequences
of applying multiple benefits tests to different choice variables with the same
optimization problem prior to moving forward with attempting to develop unit
commitment and dispatch software for application in ISO and RTO markets.

Like the first approach, this approach is fundamentally a research project whose timeline
and outcome are uncertain. This second approach has the potential to take longer than
the first approach to develop and implement ISO and RTO software capable of applying a
net benefits test based on the billing unit effect because of the initial time spent on further
research of solution methods, but it must be kept in mind that there is no assurance that
the first approach would in fact be successful. One possible outcome if the first approach
is pursued is that after several years and after spending many millions of dollars on
software development, it will become apparent that existing solution methods cannot
provide acceptable solutions to the unit commitment and dispatch problem using a net
benefits test based on the billing unit effect and that research on new solution methods
must be undertaken. A key advantage of this second approach is that it might reduce the
likelihood of such an outcome in which no solution is developed, and also offers the
potential to avoid reliance on an approach that is sub-optimal, or excessively costly to
implement. The research that would be undertaken under this approach would not only
be an alternative to pursuing the first approach but is also an alternative to proceeding
immediately to implement approaches three and four.

D. Modify Existing Software to Apply a Net Benefits Test Based on the

Billing Unit Effect on an All or Nothing Basis

A third possible approach to carrying out a dynamic net benefits test based on the billing
unit effect would be an ad hoc approach that could be carried out using only existing
industry algorithms. From a software design standpoint therefore, this approach is a
known quantity. While there would be detailed design issues to resolve in implementing

this approach, its development would be akin to normal ISO and RTO software

development projects in which these kinds of design issues are routinely resolved, tradeoffs with software performance identified and addressed in one manner or another.

In its simplest form this approach would entail a three step process; solve the dispatch
without activating demand response, regardless of whether it is economic, using the
existing dispatch and powerflow engines, then calculate clearing prices, and then solve
the dispatch again using the existing dispatch and powerflow engines but activating all
demand response that would be economic to dispatch based on its bid price and
calculating clearing prices based on this dispatch. The third step would compare
consumer payments for energy and demand response costs (but not other uplift costs that
would be borne by load),17 between the two cases and choose which solution to base
dispatch instructions or day-ahead market schedules upon, applying the net benefits test
to the two solutions.18

This approach to implementing a dynamic net benefits test based on the billing units

effect would not require development of new solution concepts but would have

implications for dispatch execution time because it would require dispatching the system to meet two distinct load levels.

If the amount of demand response whose dispatch was being evaluated was sufficiently
small, the difference between the two dispatches would be so small in terms of power
flows and constraint impacts that the second solution would likely be quite fast.19 We
have not sought to quantify exactly how small the amount of demand response would
have to be because there would not be much benefit to incurring any of these
implementation costs if the amount of demand response ultimately being dispatched was relatively small. If the amount of demand response being dispatched is not small, then
the software performance impact of a second complete20 solution could be more material. One way to address potential performance problems if there were a material number of
demand response bids would be to solve the two cases in parallel. This would somewhat increase the implementation costs and introduce some additional complexity into real-
time operations but these impacts should be manageable.

While such a three step approach to applying a dynamic net benefits test based on the
billing unit effect would likely be workable from a software implementation standpoint
(i.e. able to solve the dispatch problem within the time frame required for reliable

operation of the grid by ISOs and RTOs) in this context, it would have some limitations.
First, such a three step approach would account only for the options of dispatching all
demand response that was economic based on its bid or none of the demand response that
was economic based on its bid, which may not be the outcome the Commission intends.

Second, if there were a material amount of demand response, the activation or

disqualification of all of that demand response based on an all or nothing application of a
net benefits test based on the billing unit effect could cause ramp constraints on generator
output changes to bind in the case in which no demand response was eligible for
activation, perhaps causing a short price spike if no demand response at all were
available. If this situation arose with a net benefits test based on the billing unit effect in
place, one outcome could be that once a material amount of demand response was
activated, the activation of all of the demand response would continue to pass the net
benefits test based on the billing unit effect for some period of time, perhaps hours,
because of spurious ramp constraints that would cause price spikes to be projected if no
demand response were activated, although the activation of that demand response would
be seen to raise consumer costs if tested over a longer period of time in which the ramp
constraints ceased to bind. 21 Another possibility is that no demand response would be
activated as a result of the test and a price spike would occur because all of the demand
response would become unavailable in a single interval. These outcomes are not intrinsic
to the use of demand response nor to the application of a net benefits test based on the
billing unit effect to demand response activation but are a possible consequence of
applying an all or nothing test to all demand response activation; hence a possible
consequence of using this third approach to apply a net benefits test based on the billing
unit effect. It is possible that these kinds of suboptimal outcomes could be avoided by the
development and application of more complex ramp constraint logic in using the third
approach to apply the net benefits test based on the billing unit effect, but this is uncertain
and the impact of a more complex approach on software performance is also uncertain.

This approach could be applied in the presence of transmission congestion, although its
impact on software performance would likely be more material and might require other
changes that would raise the cost of meeting load. However, this approach could produce
a variety of unintuitive outcomes when used to apply the net benefits test based on the
billing unit effect on a congested transmission grid that might not be consistent with the
Commission’s intent.

The dynamic application of a net benefits test based on the billing unit effect utilizing this
approach on a congested transmission system would require dispatching the system with
no economic demand response dispatched, then with all economic demand response that
is in merit dispatched, and then a comparison of the net payments by remaining load
between the two cases. This net benefits calculation would be only slightly more
complex than the net benefits calculation absent congestion. However, this approach to
applying the net benefits test based on the billing unit effect has several unattractive
features in the presence of transmission congestion that might not be consistent with

FERC’s goals. First, there is a potential with this approach for the outcome of the net
benefits test for demand response at one location on the transmission system, to depend
on prices and price impacts elsewhere on the transmission system. 22 While there is an
intrinsic potential for the application of a dynamic net benefits test based on the billing
unit effect to cause price responsive loads that pay the real-time spot price to not be
dispatched off despite bids that were less than the LMP price at their location (because
the bid might not satisfy the net benefits test despite being economic), the potential lack
of predictability in the activation of individual demand response resources would likely
be exacerbated if a dynamic net benefits test based on the billing unit effect were applied
collectively on an all or nothing basis to all demand response offers across the grid of the
ISO or RTO.23

The potential for such outcomes might discourage power consumers from attempting to
respond to prices because of the unpredictability or might cause such price responsive
power consumers to adjust their consumption on their own based on real-time prices
rather than participating in ISO or RTO demand response programs, reducing the
economic benefit from their demand response because the ISO or RTO would not be able
to accurately account for their response in its unit commitment and dispatch decisions.24

Second, although it is difficult if not impossible to foresee the outcome of a dynamic net
benefits test based on the billing unit effect applied in this manner because the outcome
would depend on the amount of demand response available, its offer prices and the level
of future energy prices, applying the test across the footprint of an ISO or RTO on an all
or nothing basis, might cause the outcome of the net benefit test to vary more from
interval to interval for many resources than would be the case if the net benefits test were
individually applied to each demand response bid.25 Such an increased degree of
dispatching demand response on and off in an unpredictable manner that would be
unrelated to the bid price of the demand response resource or to LMP prices at its
location, might undermine the attractiveness of providing demand response through the
ISO or RTO dispatch. While this potential for unpredictable variability in dispatch
outcomes is intrinsic in the application of a net benefits test based on the billing unit
effect, it appears that there is a potential for this to be aggravated if this approach were
used to carry out the test on a congested transmission system.

Another limitation of this third approach would be the complexities in applying this
approach across the 24 hours of the day-ahead market. One method of isolating the
billing unit effect of activating demand response in a particular hour would be to solve
the unit commitment for the entire day, with demand response activated in a single hour
and compare the net payments by remaining load in that case to those using a unit
commitment solved with demand response not activated in any hour. This process could
then be repeated for each hour for a total of 25 solutions to the day-ahead market to apply
the net benefits test based on the billing unit effect to demand response in each hour.
This way of applying the net benefits test would, however, obviously be unworkable
from a software performance standpoint because of the number of complete day-ahead
market solutions required.

The unworkability of applying such an approach could be reduced by executing the 25

day-ahead market solutions in parallel on multiple systems, then using the results of these
solutions to apply a net benefits test based on the billing unit effect on an hour by hour
basis to determine in which hours the bids would be active. The day-ahead market would
then be solved in a final run based on the results of the net benefits test applied to these
parallel solutions. Such an approach would require extending the time frame of ISO and
RTO day-ahead markets and would entail the additional costs and implementation

complexity of solving 25 day-ahead market cases in parallel but would not require the
development and application of any new solution concepts or algorithms. However,
the results from applying such an approach might provide a very poor measure of the
actual billing unit effect if demand response passed the net benefits test based on the
billing unit effect in several hours but not in other hours, as the market solutions for the
runs which determined the activation status for demand response would potentially have
quite different unit commitments, so the results of combining the demand response
activations over the day would quite likely lead to results that were different from those
in any of the 25 cases.

Another method to account for the inter-temporal effects would be to solve the day-ahead
market twice, once with no demand response activated and once with all demand
response activated, and apply the net benefits test based on the billing unit effect to total
demand response costs and net payments by remaining load over the day, with all
demand response over the day either passing or failing the net benefits test. A limitation
of this way of solving the problem would be that demand response bids in some hours
might be cost effective yet fail the net benefits test because other demand response bids
would not be cost effective in other hours. In particular, demand response in peak hours
might not pass the test because its impact would be tested in combination with demand
response in off-peak hours that might not be cost effective as measured by the test. An
advantage of this approach beyond its simplicity is that demand response activation in the
market run would be consistent with the results of the net benefits test.

A third method of accounting for the inter-temporal impacts of demand response in

applying a net benefits test based on the billing unit effect would involve solving the day-
ahead market three times, once with no demand response activated and once with all

demand response activated.26 The net benefits test based on the billing unit effect would
then be applied to demand response in each hour separately based on the results of the
two initial runs, assuming that any difference in net payments by remaining load in a
particular hour between the two cases was a result of demand response activated in that
hour. Then the day-ahead market would be solved a third time with demand response
available for dispatch only in the hours in which it passed the net benefits test based on
the billing unit effect as applied to the two initial runs. A final step would choose the
solution with the lowest net payments by remaining load. 27 This third method of
accounting for the inter-temporal impacts avoids some of the limitations of the first and
second methods but has several limitations, a) demand response might not be activated in
a particular hour because of price increases due to changes in the unit commitment in
other hours; b) the billing unit effect in the third day-ahead market solution could be quite
different from those in the other two solutions because demand response would be turned
on in some hours but not in other hours, resulting in different unit commitments.

From a practical standpoint, the net benefits test based on the billing unit effect would need to be applied using either the second or third of the alternatives described above because of the unworkable performance impacts of the first alternative, unless it were implemented by running the 25 cases in parallel.

Overall, this third approach to applying the net benefits test based on the billing unit

effect on a dynamic basis either in the real-time dispatch or day-ahead market does not, at
least in its simplest form, require development of any new algorithms or solution
concepts, so could be implemented. The actual performance implications of this
approach on market software would need to be examined by each ISO and RTO
individually given their market design and software execution requirements and each ISO
and RTO might find it necessary to make compromises in other elements of their real-
time dispatch or day-ahead market solutions in order to accommodate the dynamic
application of the net benefits test based on the billing unit approach using this approach.

The key limitations of this approach are the consequences of the all or nothing

application of the test across demand response bids at different price levels, different
locations, and, in the context of the day-ahead market, different hours of the day.

E. Modify Existing Software to Apply a Net Benefits Test Based on the

Billing Unit Effect to Groups of Demand Response Bids.

A fourth possible approach to implementing a dynamic net benefits test based on the

billing unit effect within ISO and RTO economic dispatch programs would be structured
to take account of the possibility that if there are many demand response bids, the
activation of all of the economic demand response bids might fail a net benefits test based
on the billing units effect, but activation of a portion of the demand response bids could
pass the net benefits test. Absent development of new algorithms and solution methods

as discussed above with respect to the first approach, applying a dynamic net benefits test based on the billing unit effect that distinguishes between these two situations would entail an iterative approach in which the dispatch would be solved for incremental levels of demand response activation, with the goal of determining the optimum level of
demand response activation based on the net benefits test criterion28 within an acceptable number of iterations from a performance standpoint.

If there were a material number of demand response offers, attempting to evaluate the

billing unit effects associated with the dispatch of individual demand response bids would
likely make this fourth approach much more unworkable than the third approach from the
standpoint of dispatch execution time.29 Moreover, even if there were generally only a
small number of demand response offers, this kind of approach would still be unworkable
because ISOs and RTOs could not be assured that there would always be a small number
of demand response offers to be accounted for and would not be able to rely upon real-
time dispatch software that would sometimes solve within the required time frame and
sometimes not. 30

One way to reduce the number of incremental levels of demand response evaluations required to implement the fourth approach would be to apply the net benefits test based on the billing unit effect to bids within specified bid price ranges, rather than to each
individual demand response bid, e.g. to test all demand response bid at less than $50, then all bid at less than $55, then all bid at less than $60 etc.31

A fundamental barrier to implementing the net benefits test using these kinds of

approaches is that they would not be feasible on a congested transmission system or in the context of the unit commitment problem in the day-ahead market (or potentially in real-time dispatch software that has to solve similar non-convexities such as the
California ISO’s dispatch software).

First, it would not be possible to apply a price based grouped evaluation to demand

response bids on a congested transmission system because demand response at different

locations could have different price impacts, even if they had the same offer price. 32
While the shift factors used in congestion management would provide information
regarding the impact of demand response on flows on binding transmission constraints
that could be used in evaluating price impacts, the application of the net benefits test
based on the billing units effect on a congested system would need to account for the
differential impact on demand response at different locations on constraint shadow prices
and congestion rents, which would entail development of new solution concepts.

More generally, any grouping of demand response bids for evaluation in applying the net benefit test using existing software solution concepts would have to group them by
location, and any grouping by location would require rules for defining constrained
regions based on shift factors or congestion components, and require special rules to cover situations in which demand response activation causes transmission constraints to change from binding to non-binding or non-binding to binding.

While it would be straightforward to group bids by location on a simple radial network
with closed interface constraints, transmission congestion is not confined to closed
interfaces but typically also includes congestion on open interfaces and pre- and post
contingency line constraints. When transmission congestion is not limited to closed
interfaces demand response bids with the same price but at different locations could have
different impacts on clearing prices and very different net benefits, even if there was only
one binding transmission constraint. If ISOs and RTOs were to attempt to apply a net
benefits test based on the billing unit effect using this fourth approach and known
solution methods, they would have to apply the test using a small number of pre-defined
geographic regions to group demand response bids for evaluation, and FERC would have
to accept the fact that the groupings would often not reflect the actual congestion pattern
and that the results of the test would not always, or maybe not even often, provide even a
roughly accurate measure of the billing unit effect.

If ISOs and RTOs used say four geographic regions and four price range groupings to
simplify the application of this fourth approach, that would require 17 solutions of the
dispatch in order to apply a net benefits test based on the billing unit effect to the bid
groupings (16 cases with demand response activated in a given region for the various
price groupings and one case with none activated), then a final dispatch using the bids
that passed the test. It is hard to envision how the ISOs and RTOs could possibly solve

the dispatch 18 times within the normal time frame of the real-time dispatch even taking
advantage of every possible design change and performance improvement, so the only
way to implement a net benefits test based on the billing unit approach in this manner
would be to solve the 17 cases in parallel and feed those results into the actual dispatch
engine for the final dispatch that would determine dispatch instructions. This would be
expensive and would require a substantial time to implement in order to structure the

parallel solutions but would not require the development of any new solution concepts.33

The performance problems associated with applying fourth approach to more than a very
small number of demand response offers in the context of the dispatch problem would
become even more unworkable in the context of the unit commitment process in day-
ahead markets because of the features of the day-ahead market described above: a) the
non-convexity of the unit commitment problem, and b) interactions across the 24 hours of
the day-ahead market. It would not be workable to solve the day-ahead market for 17
different locational and pricing groupings of demand response, and do this 24 times for
each hour, requiring 408 parallel solutions. Moreover, because of the interactions across
hours and locations in the unit commitment, it is uncertain whether solving these 408
cases in parallel, then applying a net benefits test based on the billing unit effect to the
demand response in the individual hours, locations and price groups, would even produce
outcomes consistent with a net benefits test for the day as a whole.

A method that might be used to apply a net benefits test based on the billing unit effect to
groups of demand response bids in ISO and RTO day-ahead markets would be to exclude
demand response bids from the initial steps of the day-ahead market in which the unit
commitment is determined and only allow these bids to be dispatched in a final price
determination step in which the unit commitment is fixed. This approach would require
dispatching each hour 17 times to apply a net benefits test based on the billing unit effect
to 16 location/price groupings, and then solving a final dispatch based on the results of
the test, but would not involve any development of new unit commitment algorithms.

Even this approach to implementing a net benefit test based on the billing unit effect

would be unworkable in any day-ahead market in which demand response suppliers are
permitted to submit start-up bids,34 because demand response bids that are composed of
both a dispatch price and a start-up cost cannot be evaluated without taking account of
both costs, so cannot be grouped based on dispatch bids alone as required to implement
this approach.

As would be the case for the real-time dispatch, the application of the net benefits test
based on the billing unit effect to these groupings, might or might not produce a final
dispatch that satisfies the net benefits test even for demand response in aggregate, let
alone at the level of the individual bid. This potential is intrinsic in the use of
approximate locational groupings, a limited set of price groupings and applying the test
independently to each hour and could not be avoided if this fourth approach were used.
Moreover, it would not be possible to accurately assess how frequently this would be the
case until software based on the fourth approach was designed, built and placed in
operation. Once an ISO or RTO had such day-ahead market software available, it could
run test cases on historical day-ahead market data to get some sense of the quality of the
solution produced by this approach, but it likely would not be able to accurately assess
how accurately it would apply the net benefits test based on the billing unit effect until it

had actual demand response bids and observed how those bids and the software design changed the bids of other market participants, including virtual traders.

The activation of demand response after the unit commitment is fixed would also

maximize the apparent impact of the demand response activation on clearing prices in the
day-ahead market because no units could be decommitted in response to the reduced
prices. 35 However, if the generators scheduled in the day-ahead market whose operation
is made uneconomic by the activation of demand response are not committed during the
operating day and do not actually operate in real-time; this would tend to raise real-time
prices above day-ahead market prices. Such a price differential would in turn incent the
submission of additional virtual demand bids in the day-ahead market that would undo
some of the impact of the uneconomic demand response on day-ahead market prices by
driving up day-ahead market prices both in the unit commitment steps and in the final
dispatch that determines prices in combination with demand response. Hence, it could be
case that the settlement prices calculated in the pricing step of the day-ahead market
reflecting the activation of demand response would pass the net benefits test if compared
to the elevated prices in the unit commitment steps that would be driven by virtual
demand bids, yet would fail a net benefits test if compared to prices in a day-ahead
market cleared to minimize production costs.

F.Conclusions

If the first or the second approaches to implementing a dynamic net benefits test based on
the billing unit effect were taken, it would be necessary for the ISOs and RTOs to invest
some millions of dollars and perhaps a few years, or a number of years, in modifying
existing or developing new software algorithms which could consistently apply a net
benefits test based on the billing unit effect within the corresponding market run-time
window required for use in ISO and RTO real-time dispatch software and day-ahead
markets. Hence, if the first approach were taken, there could be no assurance that a
solution would be developed, and commercial software implemented, within a predefined
time frame. One possible outcome could be that after several years it would become
apparent that these methods would not produce a satisfactory solution and another
approach would have to be selected.

The only approach that could provide any assurance that a dynamic net benefits test

based on the billing unit effect could be implemented in commercial software within the
next few years would be the ad hoc third approach, in which only two levels of demand
response activation (all or none) are evaluated. Even so, the implementation of the third
approach might require simplifications in other elements of the real-time dispatch
software along with many simplifications in market design rules related to inter-temporal
constraints to maintain performance, and these simplifications may raise the actual, not

merely pecuniary, cost of meeting load. Similarly, implementation of the third approach
to applying a dynamic net benefits test based on the billing unit effect within day-ahead
markets may require simplifications in day-ahead market design that reduce the
efficiency of the day-ahead market and potentially raise the uplift costs borne by energy
consumers.

The fourth approach would, like the third approach, not involve the development of any
new software solution concepts or algorithms, but absent development of new solution
concepts, it would require very fundamental changes in the way real-time dispatch
software operates (some number of parallel solutions feeding into a final dispatch) that
would entail major changes in ISO and RTO dispatch systems. This approach would
therefore, even in the best case, require one or two more years to implement than the third
approach and its implementation complexity has the potential for operational issues to
emerge as the details are worked out that would stretch the implementation timeline even
further into the future.

IV. Other Design Choices and Market Impacts

Beyond the core design issues associated with the fundamental choice of the approach
used to develop software that would apply a net benefits test based on the billing unit
effect in ISO and RTO unit commitment and dispatch software, there are a number of
secondary design choices that would need to be made in order to utilize that software
within the existing ISO and RTO market designs and operating practices, without regard
to which of the potential approaches to implementing a dynamic net benefits test based

on the billing units effect is selected. The major issues associated with an “dynamic” net
benefits test based on the billing unit effect that would need to be addressed are listed
below:

• Application of a net benefits test based on the billing unit effect in real-time

dispatch software that includes inter-temporal optimization will require that the
Commission accept that the test results will not always be consistent with prices;

• Application of a net benefits test based on the billing unit effect will be

implemented in ways that will not be consistent with settlement prices determined by special ISO and RTO pricing rules and ISO and RTO pricing and dispatch rules will in some circumstances require that demand response resources be
dispatched to maintain reliability without regard to the outcome of a net benefits test based on the billing unit effect;

• The implementation of the net benefits test based on the billing unit effect in ISO
and RTO real-time dispatch software would not be able to account for the impact
of real-time price reductions on the day-ahead market supply curve;

• Interaction between basing activation of demand response based on a net benefits
test based on the billing unit effect and ISO and RTO market power mitigation
mechanisms could have unpredictable effects that would need to be examined

carefully and might require implementation approaches that would lead to

inconsistency between real-time prices and those used to apply the net benefit
test.

• Basing activation of real-time demand response on a net benefits test based on the
billing unit effect will impact ISO and RTO day look-ahead scheduling and unit
commitment evaluations in ways that are likely to raise the production cost of
meeting load and increase real-time price divergence between ISO and RTO
markets.

• Basing activation of real-time demand response on a net benefits test based on the
billing unit effect will hinder development of ramp management tools being
developed by some ISOs and RTOs to reliably accommodate higher levels of
intermittent generation output and will likely preclude real-time demand response
from being able to provide ramp capability in those markets if and when they are
implemented.

Each of the above listed issues is described and discussed below. A number of the issues
would impact the implementation of a net benefits calculation based on the billing unit
effect either as part of a day-ahead market and as part of the real-time dispatch, while
some others would primarily impact either day-ahead or real-time benefits calculations.

A.Interaction with real-time dispatch software that includes inter-temporal

optimization.

ISO and RTO real-time dispatch software increasingly incorporates functions that

optimize some element of the dispatch over more than one dispatch interval.36 Multi-
period optimization is currently used by New York ISO and California ISO in their real
time dispatch software (this is also sometimes referred to as “look-ahead dispatch).”
Moreover, MISO, ERCOT and ISO New England are considering implementing real-
time dispatch software with such inter-temporal optimization over the next perhaps 4-5
years. The economic benefit to power consumers from the use of inter-temporal

optimization in the real-time dispatch is that it enables the dispatch software to reduce the production cost of meeting load by accounting for known future changes in demand or supply that would require the dispatch of very high cost generation due to ramp
constraints. By dispatching low cost generation to begin moving up in advance of these changes, or high cost generation to begin moving down in advance, the cost impact of these ramp constraints can be reduced.

The kind of known future changes in demand or supply that can be taken into account in
the real-time dispatch software include projected high rates of demand increase during
portions of the early morning load pickup, large changes in scheduled inter-change at the

top of the hour, large pumps going on or off line, scheduled transmission outages that
change congestion patterns, and scheduled generation commitments or decommitments.

The ISOs and RTOs currently using or planning to implement inter-temporal

optimization in their real-time dispatch would need to apply the net benefits test based on the billing unit effect in a simplified manner that would in some circumstances lead to
inconsistencies between actual settlement prices and the outcome of the net benefits test based on the billing unit effect. The key problem is that the reason for developing
dispatch software that optimizes over time is that the optimal dispatch for the current
interval can depend on the demand and supply balance of future intervals. This in turn implies that the optimal dispatch for the current interval can depend on whether demand response will be activated in future intervals.

Projecting whether generation or demand response would be economic in future intervals
is inevitably uncertain, but it is manageable in the normal implementation of inter-
temporal optimization because if demand and prices are high in a future interval, high
cost generation or demand response would be dispatched and if prices are not high, they
would not be dispatched. Hence, if generation were dispatched up in the current interval
because high prices were projected in future intervals, those high prices would also cause
the look-ahead dispatch based on those high prices to take account of demand response
activation. However, if the activation of demand response in future periods would be
based on application of a net benefits test which accounts for the billing unit effect, there
would be no such direct link between the LMP price at the location of the demand
response resource in the future period and demand response activation, because demand
response activation would depend on the outcome of a net benefits test based on the
billing unit effect, not just on the LMP price and bid of the demand response resource.
Hence, it would be necessary for the forward looking evaluation to not only predict
whether activation in these future intervals would be economic based on the projected
LMP price at that location, but to also predict also whether activation would pass a net
benefits test based on the billing unit effect. Whether this could be accomplished within
a time frame that is acceptable to ISOs and RTOs from the standpoint of software
performance, or if could even be accomplished at all, is doubtful.

Two of the approaches discussed in section III entail the development of new software solution methods based on either known concepts or concepts yet to be developed. It is obviously not known at this point in time how those solution methods might be integrated with the inter-temporal optimization in real-time dispatch software or what compromises might be required for such an implementation.

If a net benefits test based on the billing unit effect were applied using the third approach,
i.e. collectively to all in merit demand response bids, it would similarly be possible to
apply the net benefits test based on the billing unit effect collectively to the current and
future intervals. This approach could, however, potentially lead to odd outcomes in
which demand response would be activated in the current period because of projected net
benefits in future periods or not activated in the current period because future periods
failed the net benefit test. A variation on the third approach would be to apply the net

benefits test based on the billing unit effect based on single interval dispatches with all
demand response activated or not activated, then carry out the full multi-period
optimization with demand response either activated or not activated in the initial interval
based the outcome of the net benefits test applied to the current interval, and not activated
in future intervals.37

One potential undesirable side effect of such an approach would be that the real-time
dispatch software might at times dispatch up generation out-of-merit in the current
interval to meet load in future intervals that might have been met at lower cost through
demand response activation, thus causing demand response to not be dispatched in
circumstances in which it would have been economic absent the software simplifications
needed to implement a net benefits test based on the billing units effect. In addition, there
could be situations in which demand response would pass the net benefits test based on
the billing unit effect when applied based on the single interval dispatch, but would not
have passed the test based on the settlement prices determined in the final inter-
temporally optimized dispatch. The Commission would have to accept these possibilities
if the third approach were used to implement a bet benefits test based on the billing unit
effect and applied in this manner.

Using the fourth approach to apply the net benefits test based on the billing unit effect in real-time dispatch software that optimizes over time would require solving the 17 cases,
assuming four geographic regions and four price range groupings, in parallel in a single
interval dispatch, then applying the results of those runs to determine which demand
response bids would be available in the final multi-interval dispatch. The fourth approach would have essentially the same limitations as the third approach when applied to
software that optimizes the dispatch over time as well as the limitations of the fourth
approach discussed in section III.

The central problem with implementing a net benefits test based on the billing unit effect
in this manner in real-time dispatch software that optimizes over time is that it would add
two single interval dispatch steps to software which already requires substantial
execution time in order to carry out the inter-temporal optimization. Ending use of inter-
temporal optimization by the New York ISO and California ISO and require other RTOs
to abandon their efforts to reduce the actual resource cost of meeting load and improve
reliability through improved optimization in order to implement a net benefits test
focused on achieving billing unit effects would not be a good tradeoff for power
consumers.

Hence, while inter-temporal optimization would not preclude application of a net benefit test based on the billing unit effect, FERC would have to accept the kind of
simplifications described above and the fact that the activation decision would not always be correct if evaluated against its impact on settlement prices.

B.Interaction with special RTO pricing rules

A number of ISOs and RTOs have special dispatch rules that are used to determine

settlement prices in addition to the core physical dispatch that determines base points.

Applying a net benefits test based on the billing unit effect to settlement prices calculated in accord with these special pricing rules would add complexity, slow the dispatch
execution, potentially require too many iterations to be practical and in some cases would be simply infeasible because settlement prices are calculated after the dispatch interval is over (i.e. in ex post pricing systems). These complexities would be reduced if the
Commission were to accept that the prices used to apply a net benefits test based on the billing unit effect would differ at times from actual settlement prices and accept that the net benefits test based on the billing unit effect would not be employed to determine
demand response activation in some circumstances.

Four types of special pricing rules are considered below: ex post pricing, fixed block pricing, the California ISO pricing pass and scarcity pricing.

PJM, ISO-NE, and MISO use various versions of ex post pricing systems in which

settlement prices are determined after the dispatch interval is over, using information

regarding the response of generators to their dispatch instructions, to calculate settlement
prices. It would obviously be infeasible to apply a dynamic net benefits test based on the
billing unit effect to determine demand response activation based on ex post prices that
would not be determined until after the fact. Hence, in these markets any dynamic net
benefits test based on the billing unit effect would have to be applied to prices calculated
from the ex ante dispatch instructions, which at times would turn out to be higher than
actual settlement prices.

The New York ISO has multiple pricing passes in its real-time dispatch to allow fixed
block units (such as gas turbines that are dispatched to their upper limit when operating)
to set prices in some circumstances. In order to ensure that fixed block units only set
clearing prices when they are needed to meet load, the price calculation process in the
real-time dispatch involves three distinct dispatch-passes. Moreover, in order to avoid
spurious ramp constraint driven price spikes arising from differences in generator

dispatch points among the various passes, the New York ISO dispatch has special logic to determine the application of ramp constraints across these passes.38

Beyond the adverse performance impact of solving another three pass dispatch in order to
compare prices with and without activation of demand response, we noted above with
regard to the third approach that if there were a material amount of demand response
available for dispatch the use of all on or nothing net benefits test based on the billing

unit effect to determine demand response activation could lead to short ramp constraint
driven price spikes. Any application of the net benefits test based on the billing unit
effect and utilizing the third approach would need to be tested to identify any unintended
consequences arising from interactions between the existing hybrid pricing logic
embedded in New York ISO real-time dispatch software and the way the net benefits test
would be applied on the all or nothing basis and it might be necessary for the
Commission to accept some degree of anomalies in the application of the net benefits test
based on the billing unit effect.

Similarly, under the fourth approach it would be necessary to test for interactions
between the anomalies that would inevitably arise from the individual activation of
demand response in locational and price groups and the hybrid pricing dispatch, and
again it might be necessary for FERC to accept some degree of anomalies in the
application of the net benefits test based on the filling unit effect. Until any software
developed to implement a net benefits test based on the billing unit effect and utilizing
the first or second approach was actually available, it is obviously impossible to project
what interactions might exist with the hybrid dispatch logic and what compromises might
need to be accepted in implementing such a net benefits test using those approaches.
Until this evaluation has been completed there can be no assurance that any approach
selected would generally produce accurate applications of a net benefits test based on the
billing unit effect when implemented in conjunction with the New York ISO’s current
market design. However, the alternative of eliminating the NYISO fixed block pricing
logic in order to more accurately implement a net benefits test based on the billing unit
effect would likely have serious adverse effects on the efficiency of the NYISO market.
Moreover, the impact of eliminating the hybrid pricing dispatch might be to so reduce
clearing prices that demand response would rarely if ever be economic in eastern New
York, which is likely not the outcome the Commission sought with Order 745.

Similar issues would apply to the extended LMP formulation that the Midwest ISO has
developed and filed for approval with the Commission in order to address pricing
inconsistencies that contribute substantially to excess Revenue Sufficiency Guarantee
charges. 39

The California ISO utilizes a third type of special pricing rule. The California ISO

determines physical dispatch schedules in an initial dispatch pass with very high penalty
values on some kinds of constraints to control priority of relaxation of various constraints
in cases of infeasibility, then determines settlement prices based on a second pass (pricing
run) in which constraints violated in the first pass have been relaxed. Because physical
dispatch instructions are determined in the initial physical pass (scheduling run) it would
be desirable to apply the net benefits test based on the billing units effect in this pass to
make sure that the physical dispatch accounts for demand response activation, regardless
of which approach is used to implement a net benefits test based on the billing units

effect.

It would, however, be necessary from a reliability standpoint to include an override that
would activate demand response without regard to the impact of the net benefits test if
transmission and other constraints were violated in the physical dispatch pass, to ensure
that demand response able to solve these constraints would be activated.40 Applying a
net benefits test based on the billing units effect in the physical dispatch pass could at
times cause demand response to be activated when it would not be economic or not pass
the net benefits test if the billing unit effect were calculated based on settlement prices,
which can be very different from the shadow prices in the physical dispatch pass.

A fourth type of special pricing rule is reserve shortage pricing, currently used by the

New York ISO for reserves and regulation, 41 the Midwest ISO for operating reserves, 42
ISO New England for reserve shortages, and the California ISO for operating reserve and
regulation scarcity pricing. 43 In addition, FERC has recently approved PJM’s proposal to
implement reserves shortage pricing,44 and the Midwest ISO plans to implement shortage
pricing for spinning reserves.45 A fundamental problem with applying a net benefits test
based on the billing unit effect to the activation of demand response during reserve

shortage conditions is that it would clearly be appropriate from a reliability standpoint to
activate demand response during such conditions, however there would be a potential for
activation to fail the net benefits test based on the billing unit effect if there were not
enough demand response to eliminate the reserve shortage or to materially reduce the
marginal energy offer price, so that energy prices would be set at essentially the same
level whether demand response was activated or not.46 Such an outcome in which
demand response is not activated during shortage conditions could be avoided by
implementing additional logic that would ensure that demand response would be
activated in reserves shortage conditions without regard to the outcome of a net benefits
test based on the billing unit effect.

Overall, a dynamic net benefits test based on the billing unit effect could be implemented
in combination with most ISO and RTO special pricing rules as long as the Commission
accepted that the test would have to be applied based on prices that could differ from
final settlement prices and allowed the application of a net benefits test based on the
billing unit effect to be suspended and conventional bid based production cost
minimizing benefit rules to be used in some situations, such as during reserve shortage
conditions. There might be interactions between the application of a net benefits test
based on the billing unit effect and fixed block pricing rules, particularly under software

developed under the 1st and 2nd approaches, which might be more difficult to resolve.

Whether this would turn out to be an issue could not be fully resolved until the software design for implementing a net benefits test based on the billing unit effect under one of those approaches was actually developed.

C.Accounting for the real-time price reductions on the day-ahead supply

curve

While the real-time market plays an important role in maintaining reliability and ensuring that electric load is met at least cost, very little power is typically purchased by load
serving entities at real-time prices; 5% or less of real-time load is typically settled at the real-time price. Instead, most power consumption is scheduled day-ahead and priced in the day-ahead market, and may be further hedged by bilateral long-term contracts or
vertical integration of the load serving entity into power generation.

The activation of demand response to reduce real-time spot prices does not directly

provide a billing unit effect benefit to power consumers or their load serving entities that have purchased power in the day-ahead market. Indeed, if demand response activation reduces real-time load below the load cleared in the day-ahead market and causes realtime prices to decline below day-ahead market prices, this would benefit suppliers who would be able to buy back their day-ahead positions at prices lower than their incremental costs, earning larger profits than if demand response had not been activated and they had operated their generation to meet the higher load level.

Any long run billing unit effect benefit to power consumers from paying for the

activation of demand response resources in order to depress real-time prices requires that the lower expected real-time prices attributable to demand response activation be
reflected in day-ahead market prices through additional virtual supply bid into the day-
ahead market or reduced physical load bid into the day-ahead market. However, the
introduction of additional virtual supply offered in the day-ahead market or reduced
physical load bid into the day-ahead market that reduced clearing prices in the day-ahead market would also impact the unit commitment in the day-ahead market by making it
uneconomic to commit as much generation as would otherwise have been the case. The reduced commitment of generation in the day-ahead market would reduce the billing unit effect benefit in the day-ahead market to consumers from the price depressing effects of the real-time demand response as day-ahead prices would not fall as much as would have been the case if the unit commitment were held constant.

This is illustrated in Figure 1. RS-1 portrays the real-time supply curve that would be

produced by the day-ahead unit commitment if a real-time demand of D1 were expected.
Given this supply curve and real-time demand of D1, the real-time price would be P1.
D2 portrays the real-time demand with demand response of D1-D2, and RS-2 portrays
the real-time supply curve that would be produced by the day-ahead unit commitment if a
real-time demand of D2 were expected. The introduction of demand response therefore
reduces the real-time price from P1 to P2. If one uses the real-time supply curve RS-2 to
project what prices would have been absent the demand response with a demand of D1,

the projected price would be P3, rather than the actual P1. Hence, using the real-time supply curve to apply a net benefits test based on the billing unit effect overstates the billing unit effect by P3-P1.

Figure 1

Shifts in Real-Time Supply Curves

Prices P1 and P2 are the points on the day-ahead supply curve associated with demand levels of D1 and D2. Hence, accurately measuring the billing unit effect benefit to consumers from paying for real-time demand response in order to depress real-time prices requires calculating the benefits based on the day-ahead market supply curve, rather than the real-time supply curve.

Ignoring the impact of real-time prices on day-ahead market prices and the day-ahead
unit commitment and calculating the billing unit effect benefits to consumers from real-
time demand response based on the real-time supply curve would overstate those
benefits, potentially dramatically. This is because the expectation that demand response
would be activated in real-time would reduce the amount of generation committed day-
ahead, which would raise the real-time price absent demand response activation above
what it would otherwise be. A real-time net benefits calculation utilizing the real-time
supply curve to measure the billing unit effect would be based on this inflated real-time

price, so it would overstate the day-ahead market billing unit effect benefits from realtime demand response activation.

Conversely, it would not make sense for ISOs and RTOs to expend the required resources
to implement a dynamic net benefits test based on the billing unit effect in real-time and
incur the likely adverse software performance impact if the resulting billing unit effect
benefit flowed to generators that sold power day-ahead and were dispatched down in
real-time, rather than to power consumers, but there is no practical alternative to relying
on virtual bidding to transfer the impact of real-time billing unit effect of demand
response activation to consumers buying power in the day-ahead market.

If virtual bidding is effective in transferring the impact of real-time demand response activation to day-ahead market prices, calculating the billing unit effect benefit to
consumers buying power in the day-ahead market based on the real-time supply curve will yield results that have little relationship to the actual benefits because the test applied in this would fail to account for the day-ahead market unit commitment impacts in the day-ahead market from this virtual bidding 47

We have discussed in Section III above the complexities associated with applying a
dynamic net benefits test based on the billing unit effect within the unit commitment
process of a day-ahead market. Carrying out such an analysis would be unworkable in
the context of a the real-time dispatch regardless of the approach used to implement a net
benefits test based on the billing unit effect so it would not be feasible to use such an
approach to accurately account for the billing unit effect benefits in the day-ahead market
from the impact of real-time demand response on day-ahead market prices. One possible
way to provide a roughly accurate measure of the day-ahead market impact would be to
use off-line analysis of the relationship between the slope of the real-time supply curve
and the implicit day-market supply curve to develop an adjustment factor that could be
applied to the results of a net benefits test based on the billing unit effect applied to the
real-time supply curve to roughly align the results of the calculation with the billing unit
effect benefits in the day-ahead market.48

D. Interaction with market power mitigation processes

The potential activation of economic demand response in either day-ahead or real-time
can impact the application of ISO and RTO market power mitigation by reducing the
demand that must be met with generation and capping prices. The NYISO has run day-
ahead market test cases in which a 25 megawatt reduction in demand in Zone J in the
day-ahead market caused a load pocket constraint to not bind, this in turn caused
mitigation to not be applied to certain generator offers, causing clearing prices to rise.

Similar outcomes would be possible in the Midwest ISO and ISO New England’s implementation of conduct and impact market power tests.

Similar impacts could arise with PJM’s 3 Pivotal Supplier test, the activation of demand response could reduce flows on constraints, potentially changing the outcome of the 3 Pivotal Supplier test, and causing generation offers to not be mitigated. A similar
outcome would be possible with the improved local market power mitigation process which the California ISO has recently implemented in its day-ahead market and is
working toward implementing in real-time, which is also based on a 3 Pivotal Supplier Test. We discuss the application of market power mitigation in combination with the net benefits test based on the billing unit effect below, first in the context of the real-time dispatch and then in the context of the day-ahead market.

Real-Time

Achieving consistent outcomes between the application of market power mitigation and
the actual real-time dispatch requires that the application of market power mitigation
evaluate the activation of demand response in a manner consistent with what will happen
in real-time. All ISOs and RTOs, however, presently apply their tests for real-time
market power mitigation in advance of the actual real-time dispatch. This sequencing is
important from a software performance standpoint and also to ensure that unit
commitment decisions are made based on the same bids and offers that will be used to
determine the real-time dispatch.

There is an inevitable level of uncertainty in any forward looking market power

mitigation process in that market conditions can change between the time that the market
power evaluation is carried out and the real-time dispatch, so the degree of market power
can also change. While the uncertainty associated with real-time load (and increasingly,
the uncertainty associated with intermittent generation output) is unavoidable, the errors
should not be consistently large and should not be predictable when generation offers
were submitted.

It is impossible to assess how software developed under approach one or two to apply a net benefits test based on the billing unit effect might impact the predictability of demand response activation until that software has actually been developed, If either the first or second approach were used to implement a net benefits test based on the billing unit
effect, its interaction with real-time market power mitigation designs would be one of the factors that would need to be examined under those approaches as the properties of
software developed under those approaches became known.

If the third approach were used to determine the activation of demand response in real-
time and applied the net benefits test based on the billing unit effect collectively to all
demand response bids, this could greatly increase the unpredictability of demand
response activation because that activation would be determined by an ISO or RTO wide
net benefits calculation based on the billing unit effect, rather than simply based on
whether the demand response bid price was less than the clearing price. With a net

benefits test based on the billing unit effect in operation, the forward looking evaluation used to activate market power mitigation would not only need to accurately project
market prices at the location of the demand response resources whose activation would affect the potential for the exercise of market power, it would have to accurately project prices at all locations on the grid at which prices might be impacted by demand response activation, because prices at all of these locations would impact the outcome of a net
benefits test based on the billing unit effect.

The potential for high real-time prices to result from demand response that was assumed
to be activated in evaluating the need for market power mitigation but not activated in
real-time would be limited by the fact that a substantial ISO or RTO wide price increase
resulting from the failure to apply market power mitigation in the look ahead process
would almost necessarily cause demand response activation to satisfy a net benefits test
based on the billing unit effect, in which case the activation of demand response in the
real-time dispatch would be consistent with the assumptions in the market power

evaluation. However, there may be a potential for such discrepancies between the

forward evaluation and the real-time dispatch to allow the exercise of a degree of market power in some local areas.

A particular concern would be the possibility that demand response whose activation was
projected in the forward market power evaluation to preclude the exercise of local market
power in a transmission constraint region would not be eligible for activation in real-time
because of a large amount of in merit demand response offers submitted in another region
in which their acceptance would have less impact on market prices than projected in the
forward market power evaluation, yet require substantial payments to demand response
providers, causing demand response activation to fail a net benefits test based on the

billing units effect in the real-time dispatch, allowing clearing prices to rise substantially in other transmission constrained regions when demand response is not available for activation because it fails the net benefits test.

The practical potential for such outcomes, and full assurance that there would be no
circumstance in which market power mitigation would not be triggered in the forward
evaluation because demand response was activated, yet large price increases could occur
in real-time without triggering demand response, would require careful testing of any
software based on the third approach to implement a net benefits test based on the billing
unit effect and evaluation of its performance under likely congestion patterns, price levels
and demand response offers.

The fourth approach in which the net benefits test based on the billing unit effect would
be applied to some disaggregated locational and price level groupings of demand
response bids would tend to reduce the potential for mis-forecast of prices and demand
response economics in one region to cause mis-prediction of demand response activation
in another region. However, the fourth approach could have other features that could
contribute to poor forecasts of demand response activation in the forward market power
evaluation, such as the likely need, absent development of new software algorithms, to
initialize the forward looking evaluations used to apply market power mitigation further

in advance of real-time in order to run some number of cases in parallel to determine demand response activation prior to the actual run that would determine market power mitigation, and other consequences of applying the net benefits test based on the billing unit effect to demand response bids that are roughly grouped by location and price that may not be apparent until such software can be tested. 49

Given the unpredictability of a net benefits test based on the billing unit effect and a

desire to avoid outcomes in which suppliers possessing locational market power are able
to exercise that market power because demand response is activated in the market power
test but not in the real-time dispatch, it might be that most ISOs and RTOs would choose
under any approach they ultimately use to apply a net benefits test to always evaluate the
application of market power mitigation in real-time assuming that no demand response
would be activated. If the amount of demand response available for activation is small,
this assumption would not make much difference but would result in some degree of

unnecessary mitigation if the amount of real-time demand response were material. 50
Such an approach would also be consistent with the current modeling policies of some
ISOs and RTOs who do not model load reductions from demand response activation in
forward evaluations until those load reductions are observed in real-time demand.

Day-Ahead Markets

Implementation of a net benefit test based on the billing unit effect for demand response
should not lead to unintended consequences or additional implementation issues
associated with market power mitigation in day-ahead markets using the conduct and
impact method to apply market power mitigation, but again may require that the
Commission accept that the test will not always be accurately applied. If the net benefit
test based on the billing unit effect can be applied to demand response in the initial
market based pass, the same activation outcome can be applied in the mitigated and final
passes.

For example, if the third approach to implementing a net benefits test based on the billing
unit test were used, this would entail solving the market pass with no demand response
activated, solving the market pass with all demand response activated, and then solving
the subsequent market power mitigation pass and final pass with demand response either
activated or not depending on the outcome in the market passes. Such an approach would
ensure consistency in the evaluation of market power mitigation, but could result in a

situation in which demand response would pass the net benefits test based on the billing units effect in the market pass, but would not be cost effective based on prices in the final pass if offer prices were mitigated.

This possibility could be avoided by applying the net benefit test based on the billing unit
effect to mitigated offers if mitigation were triggered, requiring that if demand response
passed the net benefit test based on the billing unit effect in the market pass and market
power mitigation were then triggered, the day-ahead market would be solved again using
mitigated prices with no demand response activated to determine if its activation would
pass the net benefits test based on the billing unit effect and mitigated prices. Solving the
day-ahead market an additional time would obviously have an additional adverse impact
on performance that would need to be addressed, potentially by simplifying other

elements of the day-ahead market design.

If the fourth approach were used to apply a net benefits test based on the billing units
effect with each locational and price based grouping of demand response bids solved in
parallel, the only apparently method of implementing this approach in the day-ahead
market would be to apply the net benefits test with the unit commitment fixed which
would also entail applying market power mitigation prior to applying a net benefit test
based on the billing unit effect to demand response offers. If this approach were taken, it
would have to be accepted that market power mitigation would be applied assuming that
there would be no demand response, potentially leading to the application of mitigation in
some instances in which it would not have been applied had the demand response offers
been taken into account.51

A net benefit test based on the billing unit effect could be applied using the third

approach to the California ISO’s day-ahead local market power mitigation in a similar
manner.52 The initial market pass would be solved both with all demand response bids
available for dispatch and with none available for dispatch. These passes would then be
used to determine the outcome of the net benefits test based on the billing unit effect
collectively for all demand response bids. Based on this determination, binding
transmission constraints would be identified based on either the case with all demand
response bids available or with none available and the three pivotal supplier test applied
to these constraints. As under a conduct and impact market power mitigation process
there would be a potential for demand response to be cost effective based on prices in the
market pass but not cost effective based on prices with mitigation applied. 53 Since the
California ISO market power mitigation evaluation is applied constraint by constraint,
running a fourth pass to apply a net benefits test based on the billing unit effect and
utilizing mitigated prices could lead to unintended outcomes if the pivotal supplier test
was applied based on the case with demand response activated and demand response then

failed the net benefit test based on the billing unit effect calculated using mitigated offer prices.54

approach were taken, it would have to be accepted that the pivotal supplier test would be applied assuming that there would be no demand response, possibly leading to occasional mitigation of offers by suppliers that would not have been mitigated had demand
response been taken into account.55

A similar approach could be used to apply the net benefit test based on the billing unit

effect in conjunction with PJM’s three pivotal supplier test in the PJM day-ahead market,
solving the market with demand response available then not available and using the
applicable case to determine constraints to which the three pivotal supplier test would be
applied. As in the California ISO market power mitigation design, there would be a
potential for demand response activation to pass the net benefits test based on the billing
unit effect utilizing prices determined in the initial unmitigated pass but to fail the test
based on mitigated prices. As with the California ISO, attempting to resolve this
inconsistency could lead to unintended consequences as well as adversely impacting day-
ahead market performance.

Hence, the net benefit test based on the billing unit effect could be applied to day-ahead market in conjunction with existing market power mitigation designs without any
additional substantial adverse impacts on market performance or risk of anomalous
outcomes, as long as the Commission is willing to allow ISOs and RTOs to apply the net benefit test utilizing unmitigated offer prices and not require that the billing unit effect be recalculated if mitigation is applied.

E.Impact on intra-day look-ahead scheduling and unit commitment

evaluations.

ISOs such as the New York ISO and California ISO that schedule imports and exports on
an economic basis intra-day (RTC and HASP) will need to account for the impact of
economic demand response in their forward scheduling evaluations to avoid inefficient
scheduling and commitment decisions that would raise consumer costs. Failure to

correctly forecast the activation of demand response could cause the New York ISO and California ISO to schedule costly imports, then dispatch down internal generation with lower offer prices in real-time when demand response is activated.

New York ISO, California ISO and PJM also use look-ahead scheduling software to

make commitment decisions for 10 and 30 minute gas turbines and some combined cycle units during the operating day and these evaluations would similarly need to account for economic demand response activation decisions in making these commitment decisions. ERCOT, ISO New England and MISO are moving towards implementation of such lookahead scheduling software over the next few years.

The ability of these look-ahead scheduling and commitment evaluations to correctly
project future demand would be adversely impacted by any additional uncertainty
regarding demand response activation that would be introduced by the net benefits test
based on the billing unit effect, just as discussed in section D with respect to market
power mitigation. It cannot be projected at this time how any implementation of the net
benefit test based on the billing unit effect developed under approaches one or two might
impact predictability, but if a dynamic net benefits test based on the billing unit effect
were implemented using the third approach, either activating all or none of the demand
response that is economic based on its bid price, this could tend to magnify the potential
prediction errors.

If the amount of economic demand response potentially activated is material, the ISOs
and RTOs would ideally want to account for it in these look-ahead scheduling decisions
so they could make cost effective short-term commitment and scheduling decisions. This
would not be important if the amount of economic demand response is not material, but if
the amount of economic demand response is not material, it would not be cost effective
for ISOs and RTOs to incur the software costs required to implement such a dynamic net
benefit test based on the billing unit effect in order to activate an immaterial amount of
demand response.

As ISO-NE and NY, NY and PJM, and PJM and MISO move to implement ISO and

RTO coordinated 15 minute interchange scheduling, all of these ISOs and RTOs will

need to account for economic demand response activation in their interchange scheduling
decisions, which will be greatly complicated and potentially unable to achieve the
intended goals if demand response activation must be determined by application of a net
benefit test based on the billing unit effect. A fundamental step in these coordinated
interchange scheduling processes is the calculation of the projected cost of interchange in
future intervals, to be used in creating supply curves that determine the least cost
interchange level. The bids of demand response resources would ideally be taken into
account in developing these supply curves. However, it is not apparent how an ISO or
RTO could determine whether to include demand response bids in the supply curve if
activation would depend on the outcome of a net benefit test based on the billing unit
effect. Attempting to calculate the outcome of a net benefit test based on the billing unit
effect for each level of interchange evaluated in constructing the forward supply curves
would make this process so complex and time consuming that it may push the time frame

for these evaluations forward in time, defeating the goal of implementing ISO and RTO
coordinated interchange in order to more closely attune schedules with current market
conditions.

The only apparent option for constructing supply curves for coordinated interchange

scheduling in these circumstances would be to simply ignore economic demand response in constructing the interchange supply curves, which would contribute to additional price divergence between the actual and projected supply curves and less optimal interchange schedules if demand response were material, reducing the benefits of implementing these coordinated interchange designs.

It would also be completely unworkable to account for the impact of activating demand response on future interchange levels in applying a net benefits test based on the billing unit effect, so the way the billing unit effect benefits would have to be calculated would tend to overstate the billing unit effect because it would not account for the increase in exports and reduction in imports associated with lower clearing prices.

Implementation of these ISO and RTO coordinated interchange scheduling designs would
be further complicated by the need to isolate the optimization function in these designs
from the calculations related to the net benefit test based on the billing unit effect. The
objective of these changes in interchange scheduling designs is to improve production
efficiency by converging prices across ISOs and RTOs, but an individual ISO or RTO
can achieve the objective of depressing energy market clearing prices by scheduling
uneconomic imports and not scheduling economic exports, actions which have a billing
unit effect benefit but tend to cause ISO and RTO prices to diverge. Unintended
interactions between the billing unit effect calculations in the net benefit test and these
interchange scheduling designs have the potential to produce unintended consequences
that could delay or even foreclose implementation of one or the other of these programs.

Moreover, if there are material amounts of economic demand response, the ISOs and RTOs will have to carefully examine the design and implementation of these forward evaluations to make sure that omitting economic demand response in forward looking evaluations does not create the potential for inefficient market participant bidding strategies designed to profit from differences between the load forecast in the forward looking evaluation and the actual real-time load.

Another complication arising with the application of a net benefit test based on the billing
unit effect to demand response would be the need to relate the unit commitment decisions
in these look-ahead unit commitment decisions to those made in the day-ahead market.
Because of the links between the day-ahead market and real-time prices that were
discussed in section C above, if ISOs and RTOs depress day-ahead market prices by
activating demand response only in the final dispatch step in the day-ahead market after
the day-ahead unit commitment is fixed, realizing the billing unit effect requires
maintaining this excess commitment in real-time.56 Maintaining this excess capacity in

the real-time dispatch and maintaining consistency between day-ahead prices and

expected real-time prices would require that ISO and RTO look-ahead evaluations in some manner accommodate the commitment of uneconomic generation to depress realtime prices in line with day-ahead market prices. While accomplishing this objective would likely not require development of new solution methods as under the first two approaches, it could require fairly significant redesign of these look-ahead commitment tools to achieve this purpose and could require that the ISOs and RTOs incur substantial costs in order to implement these software changes.

Moreover the potential complexity of developing software that would not minimize

production costs to the extent this would undo the billing rate effect of demand response activation in the day-ahead market but would somehow minimize production costs with respect to other scheduling and commitment decisions has the potential to make the
process of developing this software lengthy and expensive with the potential for surprise unintended consequences that inflict substantial costs on power consumers.

F.Interaction with Ramp Capability Products

The California ISO and Midwest ISO are currently developing ramp capability products that will entail optimizing dispatch over multiple dispatch intervals so as to maintain
ramp capability in future intervals. These forward dispatch evaluations differ from the inter-temporal optimization discussed in subsection B above in that they are designed to maintain ramp capability to respond to unforeseen demand uncertainty and supply
variability, while conventional inter-temporal optimization dispatches the system to better respond to known future demand and supply changes.

The implementation of these ramp capability products is potentially important to enabling
ISOs and RTOs to better manage of the reliability and economic impacts of increased
intermittent generation output on the transmission system. The essence of their design is
to dispatch generation slightly out of merit in order to create additional upward or
downward ramping capability when the available ramping capability would otherwise fall
below target levels.57

The implementation of a dynamic net benefit test based on the billing unit effect could

potentially hamper the implementation of these ramp capability products in two ways.

First, any adverse software performance impact from implementing a dynamic net benefit
test based on the billing unit effect using either the third or fourth approach, could make
it impossible to implement market design improvements such as these ramp capability
products, because some ISO and RTO dispatch software may not be able to execute all of
these functions within an acceptable time frame from a real-time dispatch software
performance standpoint.

virtual demand bids the day-ahead market which would tend to raise day-ahead market prices, offsetting some of the impact of the demand response.

Second, the current conceptual framework for implementing a ramp capability product,
both creating additional ramp for use in future intervals and then making that extra ramp
available for dispatch, is based on the use of penalty values for ramp that would be
incorporated in the current benefit test based on a production cost minimizing objective
function. 58 This approach would not be consistent with an objective function for demand
response activation that is based on a billing unit effect benefit calculation, yet activation
of demand response could have a material impact on the ramp needs and ramp
availability of the system. While the activation of demand response that provides
additional low cost ramp capability would likely at times also reduce energy payments to
generators, it is not clear whether that would be a general outcome, or whether activation
could instead leave the price of inflexible energy unchanged but impact the price and
amount of ramp capability on the system. On the other hand, paying LMP for demand
reductions that serve to create additional ramp could lead to high cost outcomes for
power consumers if that cost is not taken into account in the objective function in
dispatching demand response to maintain ramp.

Trying to design an ad hoc method for achieving efficient outcomes from these ramp
capability designs from the standpoint of maintaining ramp capability and maintaining
the system’s ability to respond to demand and supply shocks based on a net benefit
calculation based on the billing unit effect for demand response activation has the
potential to both greatly complicate the implementation process and give rise to
unintended consequences when the RTOs ultimately implement these designs.

V.Alternative Settlement Approach

The discussion in sections III and IV of implementation issues associated with a dynamic net benefits test based on the billing unit effect is premised on the result of the test being used to determine whether demand response that has submitted a bid price that is less
than the clearing price at its location would be dispatched. Several statements by the
Commission in orders related to Order 745 suggest that FERC does not in fact intend that demand response activation be conditioned on the outcome of the net benefits test in
addition to a demand response resource’s bid price, but only be conditioned on the
demand response resource’s bid price.

Thus, FERC stated in Paragraph 131 of Order 745A: “We clarify that pursuant to this section 206 directive, each RTO and ISO must revise its tariff to provide that when the LMP is greater than or equal to the threshold price, all demand resources that qualify for compensation will receive the LMP payment. The Commission’s section 206 action in Order No 745 did not extend, however, to situations where the LMP is not greater than or equal to the threshold price. Thus, if LMP is less than the threshold price, the Final Rule does not apply to determine the payment to a demand response resource, and any
payment will be governed by the existing RTO or ISO tariff.”

One interpretation of this statement is that the decision to activate demand response under order 745 should not depend on the outcome of the net benefit test based on the billing unit effect, and that only the level of compensation depends on the outcome of the net benefits test based on the billing unit effect.

Paragraph 127 of Order 745A similarly indicates that the activation decision would be based solely on the offer price, stating that “The Commission does not expect that a demand response provider will know the magnitude of the billing unit effect associated with its demand reduction ex ante, but if it bids its marginal opportunity cost (as we would expect in a competitive market), it will only be called when it is in the demand response provider’s economic interest to reduce consumption.”

There is similar language in the California ISO compliance filing order,59 in which

paragraph 31 states: “Order 745 did not direct RTOs and ISOs to reject demand response bids below the threshold price, nor did it determine that only bids at or above the
threshold price could result in cost-effective demand reductions. Rather, under Order No 745, when the LMP is greater than or equal to the threshold price, all demand response resources that qualify for compensation will receive the LMP payment.”60

FERC’s Order on PJM’s Compliance filing,61 and on the Midwest ISO’s compliance filing,62 contain similar statements indicating that demand response bids that do not satisfy a net benefits test based on the billing unit effect would be activated and
compensated according to relevant tariff provisions

The distinction between demand response which must be dispatched based on the

outcome of the net benefits test based on the billing unit effect and demand response

which must be compensated based on the outcome of a net benefits test based on the

billing units effect is extremely significant in the context of the workability of such a net
benefits test. If the activation of demand response in the real-time dispatch does not
depend on the outcome of a net benefits test based on the billing unit effect, there is no
need for a dynamic net benefits test based on the billing units effect in the sense of a test
carried out as part of the economic dispatch or in the process of clearing the day-ahead
market, the net benefit test based on the billing units effect could instead be carried out as
a settlement function.

Several of the implementation issues discussed in section III cease to be implementation
issues at all if the net benefits test based on the billing unit effect is to be applied after the
fact in a settlement process, while the others would be more manageable if addressed in
an after the fact settlement process. In particular, issues regarding accounting for special
ISO and RTO pricing rules, interaction with market power mitigation processes, the
impact on intra-day look ahead scheduling and commitment processes , accounting for

inter-temporal dispatch impacts, and the impact on ramp capability dispatch and scheduling algorithms would be completely avoided.

There would still be complexities associated with implementing a net benefits test based on the billing unit effect on a congested transmission grid and accounting for interactions between demand response and the unit commitment that would likely warrant some
simplifications to reduce the cost of implementing such test in the settlement system, but these complexities would not entail the development of new optimization solution
methods or algorithms. Attempting to accurately account for interactions between realtime demand response and day-ahead market prices in applying the net benefit test based on the billing unit effect would also still be complex even in an after the fact settlement process, but implementation would be more manageable in an after the fact time frame, and its implementation would not adversely impact the performance of ISO and RTO
real-time dispatch software or day-ahead markets.

Hence, it is very important to clarify what the Commission intends regarding a dynamic net benefits test based on the billing unit effect. If the Commission does not intend that the availability of demand response resources for dispatch be based on the result of a net benefits test based on the billing unit effect, but intends only that the amount of payment for demand response depend on the net benefits test based on the billing unit effect, then there is no need for this net benefits test to be carried out as a dynamic process as part of the real-time dispatch or day-ahead market. It can instead be carried out as a settlements process, in a settlements time frame, and avoid many complex interactions with elements of ISO and RTO day-ahead market and real-time dispatch.

VI.Recommendations

Five approaches for implementing order 745 have been discussed in this paper. Four of the approaches are alternative ways of applying a net benefits test based on the billing unit effect to determine the activation of demand response in the real-time dispatch or in the day-ahead market. The fifth approach would be to apply this test in a settlements
process carried out after the fact, and to base the activation of demand response bids in the real-time dispatch and day-ahead market solely on ISO and RTO market prices and the bids of demand response resources.

The second approach discussed in Section III, developing new solution methods for a net benefits test based on the billing unit effect and then developing new software based on those new methods, is the best approach from the standpoint of ultimately providing unit commitment and dispatch solutions based on the net benefits test and the billing unit
effect that could be implemented in ISO and RTO real-time dispatch and day-ahead
market software with the fewest compromises to ISO and RTO operations and the least adverse impact on the cost of meeting load. However, this approach entails an initial
research effort whose results and timeline cannot be predicted. Hence, there cannot be fixed timeline for the implementation of a net benefits test based on the billing unit effect in ISO and RTO software if this approach is selected.

The first approach discussed in Section III, developing new software applying known
methods, is also in part a research effort and therefore has an undefined timeline and
results. The implementation time line may be somewhat shorter than under the second
approach because this approach would focus on known methods, but it is not known for
sure if those methods will eventually produce a satisfactory solution in a time frame that
would be acceptable from the standpoint of ISO and RTO software performance.

The third approach, modifying existing software to apply a net benefits test based on the
billing unit effect on an all or nothing basis would likely have the shortest timeline to
implementation and the most certainty that the approach could be implemented.
However, the all or nothing application of a net benefits test based on the billing unit
effect would have an inherent potential for in merit demand response to not be eligible
for dispatch at one location because demand response offered at another location failed a
net benefit test based on the billing unit effect. Moreover, the potential unpredictability
of the outcome of a net benefit test based on the billing unit effect applied in this manner
might make it impossible to take account of demand response in many ISO and RTO
forward looking evaluations, including those used for market power mitigation,
interchange scheduling and real-time unit commitment. Hence, this approach should only
be adopted if is important to implement this aspect of Order 745 in the near term.

The fourth approach, modifying existing software to apply a net benefits test based on the
billing unit effect to groups of demand response bids would likely have a somewhat
longer timeline to implementation and more overall implementation risk than the third
approach because of the more complex software design required to implement this
approach. The advantage of this approach is that it should produce fewer anomalies in
the application of the net benefits test because of the somewhat disaggregated
application, but how material this difference would be cannot be assessed until the
software is developed and could vary between ISOs and RTOs depending on bidding and
congestion patterns.

The fifth and final approach discussed in Section VI would be to apply the net benefits test in the settlement system. This would by far be the approach with the shortest
timeline and lowest implementation cost. It would also have the least adverse impact on ISO and RTO operations, as it would not adversely impact any current or prospective look-ahead dispatch or commitment evaluations. The critical factor with this approach is simply whether it is consistent with the Commission’s intentions.

Appendix A

We begin with a simple radial example. Figure A-1 portrays the dispatch of simple

transmission system with no activation of demand response. The price of power is $130 at A, $80 at B and $25 at C.

Figure A-1

Dispatch with No Demand Response

The offers of the generation resources at each location are portrayed in Table A-2.
Table A-2

Table A-3 shows the total gross payments by load of $427,250, congestion rents of

$130,000, and total payments to generation of $297,250. Because congestion rents flow
back to power consumers directly or indirectly in the market designs of US ISOs and
RTOs and net cost to load is $297, 250, which is equal to the payments to generators.63

Table A-3

Payments by Load with no Demand Response

Gross

LoadGenerator Generator

LoadPricePaymentOutputRevenues

C105025$26,2502050$51,250

B160080$128,0002100$168,000

A2100130$273,000600$78,000

Total4750$427,2504750$297,250

Congestion Rents C-B1000$55$55,000

B-A1500$50$75,000

Total$130,000

Net Load Cost$297,250

Now suppose that there were 50 megawatts of demand response at A with a bid price of
$105 and another 50 megawatts with a bid price of $125. The demand response with a
bid price of $105 would be dispatched first, reducing load at A to 2050 megawatts and
requiring the dispatch of 550 megawatts of generation at A. Since 25 megawatts of the
generation offered at $130 would still be needed to meet load, the price of power at A
would still be $130. It would therefore be economic to dispatch another 25 megawatts of
the demand response bid in at $125, causing the price at A to fall to $125 as shown in
Figure A-4.

Figure A-4

Dispatch with Demand Response at A

The dispatch of the demand response at A would cause the price of power at A to fall by
$5 per megawatt hour (from $130 to $125 - set by the partially dispatched demand
response bid), the price paid by consumers at B and C would remain unchanged, as
would the price paid to generation at B and C, so there would be no billing unit effect
benefit to the consumers at B and C from the activation of demand response at A.
Applying the net benefits test based on the billing unit effect to the demand response at
A, the change in price at A would be -$5, reducing payments to 525 megawatts of
generation at A. Total congestion rents would also fall so the net cost of load would fall
from $297,250 to $284,875. $9750 of the reduction in load payments would be the
reduced payments by the demand response load, so the change in the cost to the
remaining load (i.e. the billing unit effect or the pecuniary benefit to power consumers
from the price reduction) would be $2625), and the payments to the demand response
would be $125 per megawatt on 75 megawatts of demand response (for a total cost of
$9375), so the payments to the demand response would considerably exceed the billing
unit effect benefits to power consumers, as shown in Table A-5.

Table A-5

Payments by Load with Demand Response at A

LoadPayment GenPayment

C10502526250205051250

B1600801280002100168000

A202512525312552565625

46754073754675284875

Benefit2025-5-10125525-2625

Cost751259375759375

Net-7506750

Alternatively, suppose that instead of the demand response at A there were 50 megawatts
of demand response offered at B, bid at $50. This demand response would be activated
and displacing 25 megawatts of generation offered at $80 and 25 megawatts offered at
$75, dropping the price at B to $75 as shown in Figure A-6. Given the transmission
congestion between A and B and between B and C, the price of power at A and C would
be unaffected by the activation of the demand response at B in this example.

Figure A-6

Dispatch with Demand Response at B

The activation of the demand response at B would pass the net benefits test. The demand response would reduce the net payments by load to $283,000 ($415,500 in gross
payments less $132,500 in congestion rents). $4,000 of the $14,250 reduction in net
payments by load would be reduced payments by demand response, so the total reduction in payments by remaining load would be $10,250.

The payments to the demand response for not consuming would be only $3750 (50

megawatts * $75 per megawatt), so the reduction in payments to generation would

greatly exceed the extra payments to the demand response providers, by $6500, as shown in table A-7.

Table A-7

Payments by Load with Demand Response at B

Gross

LoadGenerator Generator

LoadPricePayment Output Revenues

A105025$26,2502050$51,250

B155075$116,2502050$153,750

C2100130$273,000600$78,000

4700$415,5004700$283,000

Congestion Rents C-B1000$50$50,000

B-A1500$55$82,500

Total$132,500

Net Cost to Load$283,000

Original Net Cost$297,250

Reduced Payments by demand response load

at B50$80$4,000

Change in Cost to Remaining Load$10,250

Demand Response Cost50$75$3,750

Net Benefit to Remaining

Load$6,500

It is noteworthy that if the hypothesized demand response were available at both A and B
and the net benefits test were applied to combined impact of all demand response, the
demand response would fail the net benefits test. The change in cost to remaining load
would be a reduction of $12,875, but the cost of the demand response would be $13,125
($9375 + $3750) exceeding the billing unit effect by $250 shown in Table A-8.

Table A-8

Payments by Load with Demand Response at A and B

Gross

LoadGenerator Generator

LoadPricePayment Output Revenues

A105025$26,2502050$51,250

B155075$116,2502050$153,750

C2025125$253,125525$65,625

4625$395,6254625$270,625

Congestion Rents C-B1000$50$50,000

B-A1500$50$75,000

Total$125,000

Net Load Cost$270,625

Original Net Cost$297,250

Reduced payments by demand response load

at A75$130$9,750

at B50$80$4,000

Change in Gross Cost to Remaining Load$12,875

Demand Response Cost

at A75$125$9,375

at B50$75$3,750

Net Benefit to Remaining

Load-$250

Hence, in this example, portrayed in Figure A-9 below, the existence of additional

demand response at A would cause the demand response at B to not be activated if the
net benefits test based on the billing unit effect were applied jointly to all demand
response.

Figure A-9

Dispatch with Demand Response at A and B

The presence of congestion can cause activation of demand response based on a net

benefits test based on the billing unit effect applied collectively to all demand response to produce even more unintuitive outcomes with respect to demand response activation if there are changes in the level of congestion from interval to interval.

We have seen above that given the congestion pattern portrayed in Figure A-9, the

combined activation of the demand response at A and B would not pass the net benefits
test. Suppose, however, that load was lower at A so the transmission constraint between
A and B would bind at a lower shadow price absent demand response as shown in Figure
A-10 (the shadow price of the B-A constraint is $40 in Figure A-10, compared to $50 in
Figure A-1).

Figure A-10

Dispatch with No Demand Response
Lower Demand

Table A-11 shows the net cost to load would be $279,250 at this reduced level of load.

Table A-11

Payments by Load with Lower Congestion

Gross

LoadGenerator Generator

LoadPricePayment Output Revenues

C105025$26,2502050$51,250

B160080$128,0002100$168,000

A2000120$240,000500$60,000

Total4650$394,2504650$279,250

Congestion Rents C-B1000$55$55,000

B-A1500$40$60,000

Total$115,000

Net Load Cost$279,250

At the lower prices at A, only 50 megawatts of demand response at A would have bids below the clearing price and be dispatched in applying a dynamic net benefits test based on the billing unite effect as shown in Figure A-12.

Figure A-12

Dispatch with Demand Response
Lower Demand

Even with this reduced level of demand response at A and the lack of any price impact from the demand response at A, the net benefits test for demand response at A and B would be satisfied with a billing unit effect in terms of reduced payments by remaining load of $12,500 and payments to demand response providers of $9,500 ($5750 + $3750) as shown in Table A-13, producing a net benefit of $3000. Hence the demand response would be eligible to be activated and paid the LMP price at its location.

Table A-13

Payments by load with Lower Demand and Demand Response

Gross

LoadGenerator

LoadPricePayment OutputPayment

C105025$26,2502050$51,250

B155075$116,2502050$153,750

A1950115$224,250450$51,750

4550$366,7504550$256,750

Congestion Rents C-B1000$50$50,000

B-A1500$40$60,000

Total$110,000

Net Cost to Load$256,750

Original Net Cost$279,250

Reduced Payments by demand response load

at A50$120$6,000

at B50$80$4,000

Change in Cost to Remaining Load$12,500

Demand Response Cost

at A50$115$5,750

at B50$75$3,750

Net Benefit to Remaining

Load$3,000

The potential for disparate outcomes to a net benefits test in the presence of transmission
congestion as illustrated in these simple examples is inherent in applying a dynamic net
benefits test based on the billing unit effect using either of the ad hoc approaches
described in the body of this report, because either approach would entail a degree of
aggregation across demand response bids in applying the net benefits test based on the
billing unit effect.